{"$schema":"https://policywindow.org/wiki/prose-faithfulness.json","name":"Policy Window — prose-faithfulness attestation","description":"Does each instrument article's developed prose faithfully represent the primary-source law it cites? An offline adversarial judge panel (two independent judges; contradiction is refute-by-default) gave each checkable factual prose sentence a THREE-way verdict against the instrument's committed provision-excerpt corpus: supported, contradicted (a genuine defect), or out_of_corpus (the supporting text is not among the excerpts — a coverage signal, not a defect). The HEADLINE faithfulnessRate = supported / (supported + contradicted), i.e. over the ADJUDICABLE claims only; out_of_corpus is reported separately. Read deterministically at serve time. Complements /wiki/faithfulness (cell excerpt supports verdict).","docs":"https://policywindow.org/wiki/prose-faithfulness","method":"Offline adversarial judge panel (two independent judges per instrument). Each judge gives every checkable factual prose sentence a THREE-way verdict against the instrument's committed provision-excerpt corpus: supported (an on-point excerpt establishes it), contradicted (an on-point excerpt is misstated — a genuine defect), or out_of_corpus (no on-point excerpt — unadjudicable). A contradiction is refute-by-default (either judge suffices); supported requires both judges. Judgements are committed here; the serve-time report reads them deterministically (no serve-time model call).","calibrationNote":"The headline faithfulnessRate is supported / (supported + contradicted) — over the ADJUDICABLE claims only. out_of_corpus claims (whose supporting text is not among the catalog's committed excerpts — a date, a penalty figure, another law, a scholarly citation, or an un-excerpted provision) are EXCLUDED from the denominator and reported separately as a coverage signal, NOT counted against faithfulness. This keeps the rate an honest measure of whether the prose MISREPRESENTS the cited law, rather than an artifact of how much of the statute the catalog excerpts. 2026-07-02 cross-audit: the session’s later citation-inserts + 7-article deepening added checkable claims that were unjudged under the earlier headline; those 26 pending claims were re-judged on CURRENT prose by two independent judges (refute-by-default). 3 initially read as CONTRADICTED on GSA-AI-GUIDE — traced to the catalog’s own provision excerpts OVERCLAIMING the guide’s contents (SIN enumeration / sample clauses); primary-source re-verification (gsa.gov 2024-04-29) confirmed the guide is a considerations/questions resource, so the EXCERPTS were corrected and the claims re-judged (now 0 contradictions). Orphaned judgements from edited-away sentences were pruned. 2026-07-02: the 39 prior 'supported' verdicts were STRICT-re-judged under an explicit multi-limb rule (a sentence asserting several factual limbs is supported only if ALL limbs have on-point excerpts; else out_of_corpus) — 16 were multi-limb over-credits and moved to out_of_corpus, tightening the adjudicable set. faithfulnessRate remains supported/(supported+contradicted); this correction lowers the denominator, it does not inflate the rate.","auditedAt":"2026-07-02","auditVersion":"3","verifiedAt":"2026-07-01","verificationFindings":"0 missed contradictions found — the zero-contradiction result held under adversarial scrutiny. 1 over-affirmation corrected (FEDRAMP-AI-2024 supported→out_of_corpus). The 5 judge-split ‘uncertain’ verdicts are retained in the review queue (not resolved on a single auditor’s opinion).","summary":{"instrumentsWithCorpus":19,"checkableCount":170,"judgedCount":170,"pendingCount":0,"staleCount":0,"supportedCount":23,"contradictedCount":0,"outOfCorpusCount":142,"uncertainCount":5,"adjudicableCount":23,"faithfulnessRate":1,"reviewQueueCount":5},"claims":[{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Cross-Jurisdiction Position","sentence":"As the first US state statute naming \"companion chatbots\" specifically, it pioneers a use-case-targeted model contrasting with capability-tiered frameworks, addressing a relational-harm vector that broad transparency mandates leave underspecified.","judgeVerdict":"out_of_corpus","rationale":"No excerpt addresses being the first US state statute or contrasting use-case vs capability-tiered frameworks.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Key Fault Lines","sentence":"(2024) document for thin oversight (10.1145/3630106.3659051), while the § 22602(b) crisis protocol raises whether such systems act as device-like clinical decision support (10.1038/s41746-025-01544-y; Freyer et al. 2024, 10.1016/S2589-7500(24)00124-9).","judgeVerdict":"out_of_corpus","rationale":"The claim's assertion is about scholarly citations and device-like clinical-decision-support framing, which no excerpt establishes.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Operative Mechanics","sentence":"Stats. 2025, ch. 677 and codified at Cal.","judgeVerdict":"out_of_corpus","rationale":"Chaptering and codification citation details appear in no excerpt.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Implementation Trajectory","sentence":"SB 243 is in force, signed October 13, 2025, with a staged timeline: operator duties under § 22602 became operative January 1, 2026, while § 22603 annual reporting to the Office of Suicide Prevention — capturing crisis-referral data — begins July 1, 2027, deferring the empirical-accountability mechanism by eighteen months.","judgeVerdict":"out_of_corpus","rationale":"Signing/operative dates and the § 22603 reporting timeline are not represented in any excerpt.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Cross-Jurisdiction Position","sentence":"The conditional-disclosure logic of § 22602(a) echoes the EU AI Act's Article 50 transparency obligation for AI systems that interact with humans, Regulation (EU) 2024/1689, though SB 243 narrows the trigger to companion systems and to a reasonable-person misled standard.","judgeVerdict":"supported","rationale":"Excerpt 2 establishes SB 243's reasonable-person-misled trigger for the AI-notification duty that the sentence describes.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Cross-Jurisdiction Position","sentence":"SB 243 occupies a distinct niche within California's 2025 AI legislative cluster.","judgeVerdict":"out_of_corpus","rationale":"No excerpt speaks to SB 243's place within California's 2025 AI legislative cluster.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Implementation Trajectory","sentence":"That design responds to a documented gap, since freedom-of-information regimes \"generally only grant access to existing documents\" with \"no mature standard for documenting AI models\" (Olsen et al. 2024, 10.1145/3632753), making bespoke statutory disclosure the more reliable transparency channel.","judgeVerdict":"out_of_corpus","rationale":"The freedom-of-information/AI-documentation gap and its scholarship are absent from the excerpts.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Operative Mechanics","sentence":"The core duty is conditional disclosure — § 22602(a) requires a clear-and-conspicuous AI notification only \"if a reasonable person\" would be misled into believing the interlocutor is human, importing a contextual rather than per-message standard.","judgeVerdict":"supported","rationale":"Excerpt 2 verbatim establishes the conditional clear-and-conspicuous AI-notification duty triggered when a reasonable person would be misled into believing the interlocutor is human.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Key Fault Lines","sentence":"The § 22605 private right of action — injunctive relief, the greater of actual damages or $1,000 per violation, and attorney's fees — gives a remedy whose efficacy depends on contestability conditions the literature shows are easily hollowed out: studies find appeal and contestation, not nominal oversight, drive procedural fairness (10.1145/3544548.3581161), and \"meaningful\" contestation needs articulated subject needs often absent from drafting (10.1145/3757415; Alfrink et al. 2023, 10.1007/s11023-022-09611-z).","judgeVerdict":"supported","rationale":"Excerpt 3 establishes the § 22605 remedy of injunctive relief, greater of actual damages or $1,000 per violation, and attorney's fees as the sentence states.","inReviewQueue":false},{"instrument":"CA-SB-243","slug":"ca-sb-243","section":"Implementation Trajectory","sentence":"The § 22602(b) crisis protocol will likely interact with maturing post-market governance frameworks for health AI (Babic et al. 2025, 10.1038/s41746-025-01717-9) and international harmonization efforts (10.1038/s41746-025-01618-x), testing whether a use-case statute can scale into broader companion-AI safety norms.","judgeVerdict":"out_of_corpus","rationale":"The crisis-protocol excerpt does not establish the sentence's speculative assertion about interaction with health-AI post-market governance frameworks.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Implementation Trajectory","sentence":"(2025, arXiv:2502.14143) — will test whether transparency meaningfully constrains, or merely documents, frontier deployment.","judgeVerdict":"out_of_corpus","rationale":"This is a fragmentary scholarly-citation sentence about transparency documenting vs. constraining deployment; no excerpt addresses this meta-assertion.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Operative Mechanics: A Disclosure Regime Keyed to Compute","sentence":"The Attorney General enforces, with civil penalties up to $1,000,000 per violation (§ 22757.15).","judgeVerdict":"out_of_corpus","rationale":"Claims Attorney General enforcement and $1,000,000 civil penalties under §22757.15; no excerpt covers enforcement, penalties, or that section.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Operative Mechanics: A Disclosure Regime Keyed to Compute","sentence":"Its core duty is informational, not preventative — a frontier developer must publish a frontier AI framework and a pre-deployment transparency report on its website before or concurrently with release (§ 22757.12). 'Large frontier developers' (affiliate-group revenue over $500M) bear heightened duties, and § 22757.13 requires reporting 'critical safety incidents' to the Office of Emergency Services within 15 days (24 hours where danger is imminent).","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: only the §22757.12 pre-deployment transparency report limb is in corpus. The 'large frontier developer' >$500M affiliate-revenue definition and the §22757.13 15-day/24-hour incident-reporting-to-OES duty are distinct factual limbs absent from the e","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Key Fault-Lines and Critiques","sentence":"The catastrophic-risk trigger (death or serious injury to over 50 people, or over $1B in damage) privileges what Kasirzadeh (2025, 10.1007/s11098-025-02301-3) calls 'decisive' risk while neglecting 'accumulative' societal erosion.","judgeVerdict":"supported","rationale":"The §22757.11 catastrophic-risk excerpt establishes the trigger of death/serious injury to more than 50 people or over $1B in damage, matching the sentence's parenthetical description; the Kasirzadeh framing is commentary.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Implementation Trajectory","sentence":"Code § 11546.8) fall due Jan. 1, 2027, making the law's first compliance cycle a live experiment in self-described safety practice.","judgeVerdict":"out_of_corpus","rationale":"Asserts a Jan. 1, 2027 due date under Gov. Code §11546.8; no excerpt addresses dates or that code section.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Operative Mechanics: A Disclosure Regime Keyed to Compute","sentence":"Stats. 2025, ch. 138), in force since Jan. 1, 2026, regulates a narrowly drawn class: 'frontier models' trained above 10^26 FLOP, including compute used in fine-tuning (Bus. & Prof.","judgeVerdict":"supported","rationale":"The §22757.11 'frontier model' excerpt confirms the class is trained above 10^26 FLOP including compute used in fine-tuning; the in-force date and chapter citation are unaddressed but the load-bearing threshold/fine-tuning assertion is on-point and confirmed.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Cross-Jurisdiction Position","sentence":"Code § 11546.8) echoes the sovereign-compute drive that Kollar and Stokols (2025, 10.1177/0308518X251369704) trace to land, energy, and regulatory restructuring in the US and China — but as a study-and-report mandate, operative only on appropriation, not an industrial program.","judgeVerdict":"out_of_corpus","rationale":"Concerns a CalCompute study-and-report mandate under Gov. Code §11546.8 and sovereign-compute scholarship; no excerpt addresses CalCompute or that provision.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Key Fault-Lines and Critiques","sentence":"Agentic harms — a model acting 'without meaningful human oversight' or 'evading the control of its developer' — surface only via the catastrophic-risk lens (§ 22757.13), with no dedicated autonomy regime of the kind Kolt (2025, arXiv:2501.07913) and Chan et al.","judgeVerdict":"out_of_corpus","rationale":"Claims agentic harms (acting without human oversight / evading developer control) surface via §22757.13; no excerpt quotes §22757.13 or any autonomy/agentic language, so it cannot be judged from the corpus.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Cross-Jurisdiction Position","sentence":"Where the EU still wrestles with definitional instability across 'AI system, general purpose AI system, foundation model, and generative AI' (Fernández-Llorca et al. 2025, 10.1007/s10506-024-09412-y), and with foundation models that challenge 'authorship, accountability, and control' (Hulok 2025, 10.1007/s12027-025-00869-1), California sidesteps these by regulating disclosure rather than capability or output.","judgeVerdict":"out_of_corpus","rationale":"Asserts California regulates disclosure rather than capability/output vs. EU definitional debates; while the §22757.12 excerpt shows a disclosure duty, the comparative claim about sidestepping EU definitional instability is scholarly commentary not established by any excerpt.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Operative Mechanics: A Disclosure Regime Keyed to Compute","sentence":"The statute thus substitutes mandated visibility for the substantive safety mandates of its vetoed predecessor SB 1047.","judgeVerdict":"out_of_corpus","rationale":"Claims the statute substitutes mandated visibility for the substantive safety mandates of vetoed SB 1047; no excerpt addresses SB 1047 or the substitution characterization.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Key Fault-Lines and Critiques","sentence":"The 10^26 FLOP scope is the central vulnerability: Pistillo and Villalobos (2025, arXiv:2502.00003) show 'enhancement techniques' can preserve capability while cutting training compute, letting developers slip beneath the threshold — and SB 53 lacks even a standalone compute-figure report to a regulator, defining the class implicitly through § 22757.11.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] The load-bearing assertion is a negative — that SB 53 'lacks even a standalone compute-figure report to a regulator.' The corpus (which includes §22757.12's transparency-report duty) does not establish this absence, and cannot verify it. The Pistillo/Villalobo","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Implementation Trajectory","sentence":"Whether that suffices for catastrophic biosecurity or multi-agent threats — the dual-use synthesis risks mapped by Eskandar (2026, 10.1007/s43681-025-00872-9) and the miscoordination, conflict, and collusion failure modes identified by Hammond et al.","judgeVerdict":"out_of_corpus","rationale":"A fragmentary scholarly sentence about biosecurity/multi-agent threats and cited authors; no excerpt is on-point to this assertion.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Implementation Trajectory","sentence":"Core developer obligations — framework publication, transparency reports, and incident reporting under §§ 22757.12–22757.13 — became operative Jan. 1, 2026, while CalOES's annual aggregate report and the CalCompute consortium report (Gov.","judgeVerdict":"out_of_corpus","rationale":"Asserts specific operative dates (Jan. 1, 2026) for §§22757.12–22757.13 and CalOES/CalCompute reports under Gov. Code; no excerpt addresses dates or those reporting provisions.","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Cross-Jurisdiction Position","sentence":"SB 53 borrows the EU AI Act's compute-trigger logic but draws its line an order of magnitude higher: the 10^26 FLOP frontier threshold in § 22757.11 sits well above the 10^25 FLOP systemic-risk presumption of Regulation (EU) 2024/1689 (Art. 51), so the two scopes diverge rather than align.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] The sentence's core is a divergence between two figures. Only SB 53's 10^26 FLOP (§22757.11) is in this instrument's corpus; the EU AI Act Art. 51 10^25 systemic-risk presumption is a separate factual limb not in the attached SB-53 corpus. Multi-limb compariso","inReviewQueue":false},{"instrument":"CA-SB-53","slug":"ca-sb-53","section":"Key Fault-Lines and Critiques","sentence":"(2025, arXiv:2501.10114) argue agents require.","judgeVerdict":"out_of_corpus","rationale":"A fragmentary scholarly-citation sentence ('argue agents require'); no excerpt is on-point to this assertion.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Key Fault Lines: Scope Hooks, Foundation-Model Silence, and Detection Fragility","sentence":"First, the scope hook is output-and-scale, not model-level: the provisions reach a foundation-model producer only incidentally through § 22757.2–.3 output duties, never imposing a model-class obligation — leaving the 'landlords of creativity' (foundation-model providers) under-regulated, the precise gap Chau and He identify for audio deepfakes (10.1017/cfl.2025.10011).","judgeVerdict":"out_of_corpus","rationale":"The excerpts show §22757.2-.3 output duties exist, but they do not establish this claim's specific assertions about model-level scope, 'only incidentally' reach, the 'landlords of creativity' under-regulation gap, or the scholarly citation.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Cross-Jurisdiction Position: A Watermarking Convergence","sentence":"China's 2022 deep-synthesis and 2023 generative-AI rules pioneered mandatory labelling of synthetic content as a provenance model (10.1017/cfl.2024.4).","judgeVerdict":"out_of_corpus","rationale":"The claim concerns China's 2022 deep-synthesis and 2023 generative-AI labelling rules, which no CA-SB-942 excerpt addresses.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Operative Mechanics: A Provenance-and-Disclosure Triad","sentence":"Stats. 2025, ch. 674) deferred the operative date for covered-provider duties to August 2, 2026.","judgeVerdict":"out_of_corpus","rationale":"The claim is about a chaptering (ch. 674) and a deferred operative date of August 2, 2026, and none of the excerpts speak to dates or codification.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Operative Mechanics: A Provenance-and-Disclosure Triad","sentence":"Stats. 2024, ch. 291), adds §§ 22757–22757.4 to the Business and Professions Code but is adopted-not-in-force: AB 853 (Cal.","judgeVerdict":"out_of_corpus","rationale":"The claim about codifying §§22757-22757.4, adopted-not-in-force status, and AB 853 is not represented in any excerpt.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Implementation Trajectory: A Phased, Deferred Rollout","sentence":"AB 853 (§§ 22757.3.1–22757.3.3) layered new duties atop the deferred core: large online platforms and GenAI hosting platforms become operative January 1, 2027 — § 22757.3.1 barring knowing removal of provenance data and § 22757.3.2 barring hosting platforms from distributing non-disclosing systems — while capture-device manufacturers (§ 22757.3.3) follow January 1, 2028.","judgeVerdict":"out_of_corpus","rationale":"The claim concerns AB 853's §§22757.3.1-22757.3.3 platform/capture-device duties and their operative dates, none of which appear in the excerpts.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Cross-Jurisdiction Position: A Watermarking Convergence","sentence":"The EU AI Act's Article 50 imposes machine-readable marking duties on generative-AI providers and deployers; Fernández-Llorca et al. trace how that regime's underlying categories — 'AI system, general purpose AI system, foundation model, and generative AI' — remained definitionally unstable through drafting (10.1007/s10506-024-09412-y).","judgeVerdict":"out_of_corpus","rationale":"The claim is about the EU AI Act's Article 50 and its definitional instability, a separate instrument not covered by any excerpt.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Key Fault Lines: Scope Hooks, Foundation-Model Silence, and Detection Fragility","sentence":"The US patchwork Ugwuoke and Sanfilippo document (10.5325/jinfopoli.15.2025.0004) compounds this fragility.","judgeVerdict":"out_of_corpus","rationale":"The claim is a scholarly point about the US patchwork (Ugwuoke and Sanfilippo) that no excerpt addresses.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Operative Mechanics: A Provenance-and-Disclosure Triad","sentence":"Enforcement is government-only — a $5,000-per-violation civil penalty pursued by the Attorney General, a city attorney, or a county counsel (§ 22757.4), with no private right of action.","judgeVerdict":"out_of_corpus","rationale":"The claim about §22757.4's $5,000 penalty, government-only enforcement, and absence of a private right of action is not represented in any excerpt (which cover only §§22757.2-.3 duties).","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Cross-Jurisdiction Position: A Watermarking Convergence","sentence":"Crucially, empirical audit casts doubt on whether such mandates bite: Rijsbosch et al. find only 38% of image generators implement adequate watermarking and 18% deepfake labelling under the analogous EU framework (10.1002/poi3.70041), suggesting California's technical obligations may outrun present practice.","judgeVerdict":"out_of_corpus","rationale":"The claim reports empirical audit figures (Rijsbosch et al.) under the EU framework, which no excerpt addresses.","inReviewQueue":false},{"instrument":"CA-SB-942","slug":"ca-sb-942","section":"Cross-Jurisdiction Position: A Watermarking Convergence","sentence":"SB 942's latent-disclosure mandate (§ 22757.3(b)) places California within a global drift toward provenance-by-watermark, but its design diverges in instructive ways.","judgeVerdict":"uncertain","rationale":"The §22757.3 excerpt directly establishes SB 942's latent-disclosure mandate ('A covered provider shall include a latent disclosure in AI-generated image, video, or audio content'), the load-bearing factual assertion of the claim.","inReviewQueue":true},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"The Harm Landscape the Labels Target","sentence":"The Art. 17 enumeration of face generation, swap and manipulation directly targets the most acute dignitary harm — non-consensual intimate imagery — whose viral spread Kira argues existing remedies fail to contain (10.1016/j.clsr.2024.106024); the Art. 14 separate-consent requirement and the Art. 12 complaint channel are the Provisions' principal redress hooks for affected individuals.","judgeVerdict":"supported","rationale":"Excerpt [5] enumerates face generation/swap/manipulation under Art. 17, excerpt [2] establishes the Art. 14 separate-consent requirement for biometric editing, and excerpt [0] establishes the Art. 12 complaint channel — all three redress hooks are on-point; the scholarly-harm citation is separate and unaddressed but the statutory assertions hold.","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Cross-Jurisdiction Position","sentence":"The Art. 17 conspicuous-labelling rule for face- and voice-synthesis services parallels the transparency obligation in Regulation (EU) 2024/1689 (Art. 50 of the AI Act), but where the EU instrument turns on a harmonised statutory definition of 'deep fake' whose teleology Abuz analyses (10.1002/poi3.435), the Chinese Provisions define the broader category of 'deep synthesis' functionally in Art. 23 and reach all generative text, audio, image and scene services, not only deceptive ones.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] The load-bearing claim that the Provisions 'define the broader category of deep synthesis functionally in Art. 23' rests on Art. 23, which is not in corpus; nor is the 'reaches all generative text/audio/image/scene' scope. The EU Art. 50 parallel and Abuz anal","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Cross-Jurisdiction Position","sentence":"They are also the cross-referenced predecessor to China's 2023 Interim Measures for Generative AI Services and its 2025 synthetic-content labelling standard (White & Case 2025).","judgeVerdict":"out_of_corpus","rationale":"The core assertion is a predecessor/cross-reference lineage to China's 2023 Interim Measures for Generative AI Services and its 2025 synthetic-content labelling standard. No committed excerpt (redress, training data, biometric, transparency, provenance, deepfakes, national-security) speaks to that instrument-lineage claim; it is un-adjudicable from the provision corpus. | The sentence asserts thes","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Key Fault-Lines and Critiques","sentence":"A further fault-line is scope and enforcement: the Provisions bind providers and technical supporters operating within mainland China, leaving cross-border synthetic media and individual bad actors largely beyond reach, and the consent rule of Art. 14 does little for victims of content produced abroad.","judgeVerdict":"out_of_corpus","rationale":"No excerpt addresses the territorial binding of providers/technical supporters to mainland China or cross-border reach; excerpt [2]'s Art. 14 consent rule is mentioned but the claim's assertion about geographic scope/enforcement limits is not represented.","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"The Harm Landscape the Labels Target","sentence":"The labelling mandate is best read as a response to the distinctive epistemic and dignitary harms of synthetic media.","judgeVerdict":"out_of_corpus","rationale":"This is an interpretive framing about the labelling mandate responding to epistemic/dignitary harms of synthetic media — no excerpt speaks to the mandate's purpose or rationale.","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Key Fault-Lines and Critiques","sentence":"Detection on the consumption side is no panacea either: Harris catalogues the shortcomings of AI deepfake detectors (10.1007/s13347-024-00700-8), and Groh and colleagues show that even human detection of political-speech deepfakes is unreliable across modalities (10.1038/s41467-024-51998-z) — so a conspicuous-label requirement (Art. 17) shifts the burden onto a verification ecosystem that does not robustly exist.","judgeVerdict":"out_of_corpus","rationale":"The claim rests on scholarly findings about deepfake-detector shortcomings and human-detection unreliability; while it references the Art. 17 conspicuous-label requirement (which excerpt [5] supports), the load-bearing assertion about verification-ecosystem inadequacy is not in the corpus.","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Key Fault-Lines and Critiques","sentence":"Zhang and colleagues prove that strong watermarking of generative outputs is, under broad assumptions, impossible against a motivated adversary (arXiv:2311.04378), so the Art. 16 embedded-identifier duty and the Art. 18 anti-removal prohibition may not survive determined circumvention.","judgeVerdict":"uncertain","rationale":"The claim's assertion is a scholarly impossibility result about watermarking against adversaries; the Art. 16 identifier duty and Art. 18 anti-removal prohibition are named (excerpts [3],[4] support their existence) but the claim's substance about circumvention survivability is not established by any excerpt.","inReviewQueue":true},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Key Fault-Lines and Critiques","sentence":"The regime's central limitation is the technical fragility of the labels it mandates.","judgeVerdict":"out_of_corpus","rationale":"An interpretive conclusion that the regime's central limitation is the technical fragility of its labels — no excerpt addresses label fragility or characterizes the regime's limitations.","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Operative Mechanics: A Two-Tier Labelling Regime","sentence":"The heightened duty (Art. 17) requires conspicuous, public-facing labels on services that could confuse or mislead, expressly enumerating face generation, face swapping, face manipulation and pose control among regulated image and video editing.","judgeVerdict":"supported","rationale":"Excerpt [5] establishes the Art. 17 conspicuous public-facing labelling duty and expressly enumerates face generation, face swap, face manipulation and pose control among regulated image/video services — directly matching the claim.","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Operative Mechanics: A Two-Tier Labelling Regime","sentence":"Provider obligations extend beyond labelling: Art. 14 requires the separate consent of an individual whose facial or vocal biometric information is edited; Art. 9 mandates real-identity verification of users; Art. 12 requires accessible complaint and public-reporting channels; and Arts. 19-20 impose algorithm filing and security assessment for services with public-opinion or social-mobilisation attributes.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: Art. 14 separate-consent and Art. 12 complaint channels are covered, and Arts. 19-20 filing/security-assessment appear in the Art. 6 summary excerpt, but the Art. 9 real-identity-verification-of-users limb is a distinct factual assertion absent fro","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Cross-Jurisdiction Position","sentence":"Compared with the liability-and-privacy lattice that Novelli and colleagues map across EU law for generative AI (10.1016/j.clsr.2024.106066), the Provisions are administratively front-loaded: filing (Art. 19), security assessment (Art. 20) and real-name verification (Art. 9) operate ex ante through the regulator rather than ex post through courts.","judgeVerdict":"uncertain","rationale":"Excerpt [6] names Art. 19 filing and Art. 20 security assessment, but no excerpt establishes Art. 9 real-name verification nor the ex-ante-through-regulator vs ex-post-through-courts characterization that is the claim's core comparative assertion.","inReviewQueue":true},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Operative Mechanics: A Two-Tier Labelling Regime","sentence":"The baseline duty (Art. 16) requires providers to embed technical identifiers — implicit, watermark-type markers that do not impede a user's experience — in all generated or edited content, while Art. 18 prohibits any party from deleting, altering or concealing those markers.","judgeVerdict":"supported","rationale":"Excerpt [3] establishes the Art. 16 duty to add technical identifiers that do not affect user use to generated/edited content, and excerpt [4] establishes the Art. 18 prohibition on deleting, tampering with, or concealing those identifiers — both parts of the claim are on-point and accurate.","inReviewQueue":false},{"instrument":"CN-DEEPSYN-2022","slug":"china-deep-synthesis-provisions","section":"Operative Mechanics: A Two-Tier Labelling Regime","sentence":"The Provisions (CAC Order No. 12, effective 10 January 2023) build a two-tier labelling regime over 'deep synthesis' services.","judgeVerdict":"out_of_corpus","rationale":"No excerpt states the CAC Order No. 12 designation, the 10 January 2023 effective date, or characterizes the regime as 'two-tier'; these are metadata/interpretive assertions absent from the corpus.","inReviewQueue":false},{"instrument":"DFARS-252-204","slug":"dfars-252-204","section":"AI Coverage as a Designation Artefact, Not a Technology Rule","sentence":"The data-leakage concern is acute precisely because, as Ruschemeier shows, generative models \"memorize and leak pieces of training data\" and so resist treatment as anonymous (10.1017/cfl.2024.2).","judgeVerdict":"out_of_corpus","rationale":"The claim concerns generative models memorizing/leaking training data (Ruschemeier); neither excerpt addresses generative models, training data, or anonymity, so it cannot be judged from the corpus.","inReviewQueue":false},{"instrument":"DFARS-252-204","slug":"dfars-252-204","section":"The Subpart as a National-Security Overlay Regime","sentence":"Where EU instruments bolt security exemptions onto a general AI law — producing what Palmiotto calls \"double standards for fundamental rights protection\" (10.1017/err.2024.97) and what Yazici flags as military/defence exclusions leaving surveillance under-regulated (10.1080/17579961.2025.2470589) — DFARS is itself the security-specific floor, not an exception carved from a broader regime.","judgeVerdict":"out_of_corpus","rationale":"The comparative framing of EU security exemptions (Palmiotto, Yazici) versus DFARS as a 'security-specific floor' is not addressed by either excerpt, which only state the security and reporting duties without any comparative characterization.","inReviewQueue":false},{"instrument":"DFARS-252-204","slug":"dfars-252-204","section":"Operative Mechanics: CDI Safeguarding and 72-Hour Incident Reporting","sentence":"Clause 252.204-7012(c) adds the reactive obligation, requiring contractors to \"rapidly report cyber incidents to DoD … within 72 hours of discovery.\" The 2020 CMMC interim rule layered attestation on top via 252.204-7019/-7020/-7021, scaling third-party certification of 800-171 implementation by contract tier — a posture that treats verifiable inputs as the governance lever, echoing arguments that compute is uniquely regulable because it is \"detectable, excludable, and quantifiable\" (Sastry et al. 2024, arXiv:2402.08797) and that input metrics work best to flag, not to score, risk (Heim & Koessler 2024, arXiv:2405.10799).","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: only the 252.204-7012(c) 72-hour reporting duty is in corpus. The 2020 CMMC interim-rule attestation via 252.204-7019/-7020/-7021 and tiered third-party certification are a distinct factual limb absent from the excerpts (which contain only 7012(b) ","inReviewQueue":false},{"instrument":"DFARS-252-204","slug":"dfars-252-204","section":"AI Coverage as a Designation Artefact, Not a Technology Rule","sentence":"This is a structurally different posture from purpose-built AI law, where autonomous content generation \"challenges legal categories of authorship, accountability\" and forces bespoke risk tiers (Hulok 2025, 10.1007/s12027-025-00869-1).","judgeVerdict":"out_of_corpus","rationale":"The claim about purpose-built AI law, autonomous content generation, and bespoke risk tiers (Hulok) is not the subject of either excerpt, so no on-point text exists in the corpus.","inReviewQueue":false},{"instrument":"DFARS-252-204","slug":"dfars-252-204","section":"The Subpart as a National-Security Overlay Regime","sentence":"Statewatch's account of exemptions making supervision \"extremely difficult\" (Jones and Lanneau 2025) describes the rights-displacement risk; the DFARS analogue is opacity through classification, where CDI designation can shield AI systems from the transparency a civil instrument would demand.","judgeVerdict":"out_of_corpus","rationale":"The assertion about CDI designation, classification-based opacity, and shielding AI systems is not represented in either excerpt, which say nothing about classification or transparency-shielding.","inReviewQueue":false},{"instrument":"DFARS-252-204","slug":"dfars-252-204","section":"AI Coverage as a Designation Artefact, Not a Technology Rule","sentence":"Yet DFARS regulates the system-of-record, not the model's emergent disclosure behaviour — a spilled training corpus is reportable, but inference-time leakage from a deployed weight set sits awkwardly outside the incident taxonomy the clause was drafted around, a generative-AI cybersecurity gap also flagged in EU-law analyses (Novelli et al. 2024, 10.1016/j.clsr.2024.106066).","judgeVerdict":"out_of_corpus","rationale":"The claim's assertions about system-of-record versus emergent disclosure, inference-time leakage, and the incident taxonomy's scope go beyond the excerpts; the reporting excerpt does not speak to what falls inside or outside its taxonomy, so the point cannot be judged.","inReviewQueue":false},{"instrument":"DFARS-252-204","slug":"dfars-252-204","section":"Operative Mechanics: CDI Safeguarding and 72-Hour Incident Reporting","sentence":"Clause 252.204-7012(b) imposes the substantive duty: contractors must \"provide adequate security on all covered contractor information systems … by implementing NIST Special Publication 800-171\" (revision 2).","judgeVerdict":"supported","rationale":"The quoted substantive duty to 'provide adequate security on all covered contractor information systems … by implementing NIST Special Publication 800-171' matches excerpt 252.204-7012(b) verbatim and is on-point (the '(revision 2)' parenthetical is extra but the excerpt specifies no conflicting revision).","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"Adoption Trajectory and Outlook","sentence":"Since adoption the S&IP has functioned as the connective tissue between the 2020 Ethical Principles and operational practice, with the CDAO consolidating governance authority and maintaining the RAI Toolkit as the living delivery vehicle — meaning the instrument's real bite evolves through toolkit revision rather than amendment of the strategy text.","judgeVerdict":"out_of_corpus","rationale":"The excerpts state only the four Ethical Principles; nothing addresses the S&IP as 'connective tissue', the CDAO consolidating governance, or the RAI Toolkit, so no excerpt is on-point.","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"Critiques and Coverage Gaps","sentence":"Foundation-model-specific duties are not addressed directly; they 'flow through' Toolkit guidance regardless of architecture — a gap sharpened by definitional instability in the very category (Fernandez-Llorca et al. 2025, 10.1007/s10506-024-09412-y) and by the fact that autonomous generation strains legal categories of authorship, accountability, and control that any procurement assurance must rest on (Hulok 2025, 10.1007/s12027-025-00869-1), compounded by training-data leakage risks (Ruschemeier 2025, 10.1017/cfl.2024.2).","judgeVerdict":"out_of_corpus","rationale":"No excerpt mentions foundation models, Toolkit guidance flow-through, or the cited scholarship; the assertion is not represented in the four principle excerpts.","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"Critiques and Coverage Gaps","sentence":"Compute-threshold reporting is likewise implicit, surfacing only through standard acquisition channels, even as scholarship shows such thresholds are vulnerable to compute-reducing enhancement techniques (Pistillo & Villalobos 2025, arXiv:2502.00003).","judgeVerdict":"out_of_corpus","rationale":"Nothing in the excerpts concerns compute-threshold reporting or acquisition channels, so this claim cannot be judged from the corpus.","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"Critiques and Coverage Gaps","sentence":"Most acutely, redress under 'Governable' reaches only operator-facing disengagement, not affected-civilian remedy — a deficit relative to empirically grounded contestability needs (Yurrita et al. 2025, 10.1145/3757415) and the explainability-contestability link for public-sector AI (Schmude et al. 2025, arXiv:2504.18236).","judgeVerdict":"supported","rationale":"The 'Governable' excerpt indeed limits itself to detecting/avoiding unintended consequences and disengaging/deactivating deployed systems (operator-facing disengagement), supporting the claim that 'Governable' reaches only operator-facing disengagement and not affected-civilian remedy.","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"What the Strategy Commits To","sentence":"The 'Reliable' principle commits capabilities to 'explicit, well-defined uses' subject to testing within those uses, while 'Traceable' commits to documentation and explainability so that relevant personnel understand the technology, its development, and its operational methods — an obligation that maps onto the explainability needs scholarship identifies for accountable public-sector AI (Schmude et al. 2025, arXiv:2504.18236).","judgeVerdict":"supported","rationale":"The 'Reliable' excerpt ('explicit, well-defined uses' subject to testing within those uses) and the 'Traceable' excerpt (relevant personnel possess appropriate understanding of the technology, development processes, and operational methods) directly establish both halves of this sentence.","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"Standing Relative to Binding Law","sentence":"It is deliberately distinct from DoDD 3000.09 (Autonomy in Weapon Systems), the directive that governs lethal-autonomy decisions and was separately updated in January 2023; the S&IP routes catastrophic and mission risk through the 'Reliable' principle and JCIDS validation while leaving LAWS-specific decisions to that directive (U.S.","judgeVerdict":"out_of_corpus","rationale":"Multi-limb claim about distinctness from DoDD 3000.09, its Jan-2023 update, LAWS routing, and JCIDS validation. Only the 'Reliable' principle limb has any excerpt (testing/assurance), but the sentence's core (separate directive, LAWS, JCIDS) is external and un-excerpted, so the on-point core is only partially adjudicable. | Multi-limb claim about the S&IP being distinct from DoDD 3000.09, that dir","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"Adoption Trajectory and Outlook","sentence":"Its catastrophic-risk handling, channeled through the 'Reliable' principle and JCIDS, remains partial against scholarship urging precautionary state obligations to regulate AI's extreme tail risks (Druzin et al. 2025) and the distinction between decisive and accumulative existential risk (Kasirzadeh 2025, 10.1007/s11098-025-02301-3); nuclear-domain AI dynamics further test its mission-bounded frame (Allison & Herzog 2025, 10.1111/risa.70105).","judgeVerdict":"out_of_corpus","rationale":"The excerpts do not discuss catastrophic-risk handling via JCIDS or any existential-risk scholarship; the 'Reliable' excerpt does not address risk channeling, so it is not on-point.","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"Adoption Trajectory and Outlook","sentence":"As a voluntary internal pathway it cannot supply the binding audit-and-halt machinery proposed for international AI safety governance (Scholefield et al. 2025, arXiv:2503.18956), leaving its trajectory dependent on continued executive commitment.","judgeVerdict":"out_of_corpus","rationale":"Nothing in the excerpts characterizes the instrument as a 'voluntary internal pathway' lacking binding audit-and-halt machinery; that assertion is outside the quoted principle text.","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"Standing Relative to Binding Law","sentence":"This self-regulatory posture exemplifies the national-security pattern that civilian regimes externalize: where the EU AI Act, Regulation (EU) 2024/1689, embeds security exemptions that critics argue make oversight 'extremely difficult' (Jones & Lanneau 2025) — exceptions that widened through negotiation to produce 'double standards for fundamental rights protection' (Palmiotto 2025, 10.1017/err.2024.97) — the DoD instead writes its own internal framework rather than carving out from a binding civilian baseline.","judgeVerdict":"out_of_corpus","rationale":"The excerpts contain only the ethical principles and say nothing about a self-regulatory posture, the EU AI Act, or DoD writing its own framework versus a civilian baseline.","inReviewQueue":false},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"What the Strategy Commits To","sentence":"The June 22, 2022 Strategy and Implementation Pathway (S&IP) is an internal DoD policy statement, not a statute, that operationalizes the five Ethical Principles for AI adopted Feb. 24, 2020 — Responsible, Equitable, Traceable, Reliable, Governable.","judgeVerdict":"uncertain","rationale":"The four excerpts are the Ethical Principles Governable, Reliable, Responsible, and Traceable, corroborating the named principles; the excerpts confirm these are DoD principles for AI (the date and 'not a statute' framing are extra but the on-point principle-naming assertion is established).","inReviewQueue":true},{"instrument":"DOD-RAI-2022","slug":"dod-rai-strategy","section":"What the Strategy Commits To","sentence":"It does so through six foundational tenets: RAI governance clarifying roles among OUSD(R&E), OUSD(A&S), the DoD CIO and the CDAO; warfighter trust via calibrated reliance and T&E/V&V — the kind of assurance machinery that broader governance scholarship warns is still underdeveloped, lacking 'the mechanisms and institutions to prevent misuse and recklessness' (Bengio et al. 2024, 10.1126/science.adn0117); integration of RAI across the acquisition lifecycle; requirements validation through JCIDS gating; a responsible AI ecosystem covering data sourcing and vendor disclosure; and an AI workforce training mandate.","judgeVerdict":"out_of_corpus","rationale":"The excerpts do not enumerate the six foundational tenets (RAI governance, warfighter trust, acquisition integration, JCIDS, ecosystem, workforce) or the cited scholarship; these are not represented in the principle excerpts.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Key fault lines: enforcement architecture and rights redress","sentence":"Ho-Dac, arXiv:2402.16869), and whether maximum harmonisation pre-empts legitimate national AI policy (Veale & Zuiderveen Borgesius 2021).","judgeVerdict":"out_of_corpus","rationale":"Concerns maximum-harmonisation pre-emption and a scholarly citation; no excerpt addresses harmonisation/pre-emption of national policy.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Cross-jurisdiction position: the only binding horizontal regime","sentence":"(vol. 63); Migliorini, doi:10.1016/j.clsr.2024.105985).","judgeVerdict":"out_of_corpus","rationale":"Bare scholarly citation fragment with no statutory assertion any excerpt can bear on.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Implementation and trajectory: staggered timeline and the Digital Omnibus","sentence":"A provisional inter-institutional agreement reached on 7 May 2026 deferred the high-risk deadlines — Annex III obligations from 2 August 2026 to 2 December 2027 (a 16-month slip tied to standards availability), and Annex I obligations from 2 August 2027 to 2 August 2028 — while adding targeted measures such as a ban on 'nudifier' applications (European Parliament press release, 23 March 2026).","judgeVerdict":"out_of_corpus","rationale":"Asserts specific 2026 deferral dates and a nudifier ban; no excerpt covers deadlines, deferrals, or nudifier applications.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Operative mechanics: a risk-tiered, product-safety architecture","sentence":"Deployers face a lighter but distinct set, including human oversight in operation (Art. 26(2)), log retention (Art. 26(6)), worker information (Art. 26(7)), and — for public bodies and certain private deployers — a Fundamental Rights Impact Assessment (Art. 27).","judgeVerdict":"out_of_corpus","rationale":"The only Art. 26 excerpt is the general 'use per instructions' duty; it does not establish the specific sub-duties (26(2)/26(6)/26(7)) or Art. 27 FRIA, nor contradict them.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Operative mechanics: a risk-tiered, product-safety architecture","sentence":"A separate transparency tier (Art. 50) requires disclosure for chatbots, deepfakes, and synthetic-media labelling — though an empirical audit of generative tools finds adoption of these labelling duties remains partial, with only 38% implementing adequate watermarking and 18% deepfake labelling (Rijsbosch et al. 2026, doi:10.1002/poi3.70041).","judgeVerdict":"supported","rationale":"Art. 50 excerpts directly establish transparency/disclosure for deepfakes, synthetic-media marking, and chatbot AI-interaction, which the sentence's core assertion tracks (the 38%/18% figure is extra-corpus but the on-point core is supported).","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Key fault lines: enforcement architecture and rights redress","sentence":"Commentators also flag the self-assessment default for most Annex III systems (Art. 43 permits internal conformity assessment for many categories, reserving third-party bodies for biometrics), the breadth and contestability of the Art. 51 systemic-risk presumption, and the dependence of the entire high-risk regime on harmonised standards that were not finalised on schedule.","judgeVerdict":"out_of_corpus","rationale":"Turns on Art. 43 internal-vs-third-party conformity assessment and standards timing; no excerpt covers Art. 43 conformity mechanics or standards scheduling.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Implementation and trajectory: staggered timeline and the Digital Omnibus","sentence":"(EU) 2024/1689 on 12 July 2024) and applies on a staggered schedule under Art. 113: prohibited practices (Art. 5) from 2 February 2025, GPAI obligations (Chapter V) from 2 August 2025, and high-risk obligations from 2 August 2026 (Annex III use-cases) and 2 August 2027 (Annex I product-regulated systems).","judgeVerdict":"out_of_corpus","rationale":"Asserts specific Art. 113 staggered application dates; no excerpt establishes any dates.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Key fault lines: enforcement architecture and rights redress","sentence":"Veale and Zuiderveen Borgesius argue the draft leaned on '1980s product safety regulation' and delegated substantive content to standardisation bodies 'with no fundamental rights experience', warning that key protections turn on essential-requirements text operationalised through CEN-CENELEC harmonised standards rather than legislative specification (arXiv:2107.03721; 22 Computer Law Review International 97 (2021)).","judgeVerdict":"out_of_corpus","rationale":"Scholarly critique of product-safety framing and CEN-CENELEC standardisation; no excerpt addresses standardisation bodies or essential-requirements delegation.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Implementation and trajectory: staggered timeline and the Digital Omnibus","sentence":"By late 2025 the standards pipeline and supporting tools were visibly behind schedule, prompting the Commission to table the 'Digital Omnibus on AI' on 19 November 2025 as part of a broader simplification drive (European Commission 2025).","judgeVerdict":"out_of_corpus","rationale":"Assertion is about the 2025 standards pipeline being behind schedule and the Commission tabling the 'Digital Omnibus on AI' on 19 Nov 2025. No committed provision excerpt covers standards timing or the Digital Omnibus; a date/event claim with no on-point excerpt. | The sentence concerns the standards pipeline being behind schedule and the Commission tabling the 'Digital Omnibus on AI' on 19 Novemb","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Cross-jurisdiction position: the only binding horizontal regime","sentence":"The Council of Europe's Framework Convention on AI (CETS No. 225, opened for signature 5 September 2024) is binding by ratification but principles-based and rights-focused, lacking the Act's granular conformity machinery (Council of Europe 2024, CETS No. 225).","judgeVerdict":"out_of_corpus","rationale":"Core is a comparison to the Council of Europe Framework Convention (CETS 225) — its binding-by-ratification, principles-based, rights-focused character. No excerpt establishes the convention or the Act's 'granular conformity machinery' contrast; external-instrument claim, un-adjudicable. | Comparison to the Council of Europe Framework Convention on AI (CETS No. 225), its opening for signature, and","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Operative mechanics: a risk-tiered, product-safety architecture","sentence":"Regulation (EU) 2024/1689 structures obligations around an escalating risk taxonomy rather than a sectoral or technology-specific frame.","judgeVerdict":"out_of_corpus","rationale":"High-level framing that obligations use an escalating risk taxonomy 'rather than a sectoral or technology-specific frame'; no single excerpt establishes this structural comparison.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Key fault lines: enforcement architecture and rights redress","sentence":"A related concern is the redress gap: the original proposal offered affected persons no individual complaint or judicial remedy, a deficiency only partially addressed in the final text via the Art. 85 right to lodge complaints with a market-surveillance authority and the Art. 86 right to explanation of individual decisions — remedies critics still view as thin relative to GDPR-style enforcement.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: the Art. 85 market-surveillance complaint right is in corpus, but the Art. 86 right to explanation of individual decisions is a distinct factual limb not in the excerpts, and the 'original proposal offered no remedy' historical claim is likewise un","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Operative mechanics: a risk-tiered, product-safety architecture","sentence":"General-purpose AI (GPAI) is governed by a parallel Chapter V regime: baseline documentation and copyright/training-data obligations (Art. 53) escalate to model evaluation, systemic-risk assessment, adversarial testing, incident reporting, and cybersecurity (Art. 55) once a model is classified as posing 'systemic risk' under Art. 51 — presumptively triggered when cumulative training compute exceeds 10^25 floating-point operations (Art. 51(2)), a bright-line whose robustness is contested given enhancement techniques that cut measured training compute while preserving capability (Pistillo & Villalobos 2025, arXiv:2502.00003).","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: Art. 53 training-data documentation and the Art. 51 10^25 FLOP systemic-risk presumption are covered, but the Art. 55 escalation duties (model evaluation, adversarial testing, incident reporting, cybersecurity) are a distinct factual limb absent fr","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Implementation and trajectory: staggered timeline and the Digital Omnibus","sentence":"Early enforcement has nonetheless begun under the in-force prohibition and GPAI tiers, with reported market-surveillance scrutiny of large platforms (Council of the EU press release, 7 May 2026).","judgeVerdict":"out_of_corpus","rationale":"Asserts early enforcement/market-surveillance activity tied to a May 2026 press release; no excerpt addresses enforcement activity or timing.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Operative mechanics: a risk-tiered, product-safety architecture","sentence":"The bulk of the regime governs 'high-risk' systems, classified via Art. 6 by reference to Annex I product-safety legislation and the Annex III use-case list (e.g. biometrics, critical infrastructure, employment, essential services).","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] The load-bearing classification mechanism — 'via Art. 6 by reference to Annex I product-safety legislation and the Annex III use-case list' — rests on Art. 6 and Annex I, neither of which is in corpus. Annex III examples are partly present, but the core Art. 6","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Cross-jurisdiction position: the only binding horizontal regime","sentence":"The United States lacks a federal counterpart: a market-led posture was reinforced when the January 2025 executive order rescinded prior safeguards, and the December 2025 'national policy framework' order directed agencies to contest divergent state laws (90 Fed.","judgeVerdict":"out_of_corpus","rationale":"Claim is about the US lacking a federal counterpart and about January-2025 and December-2025 US executive orders. No committed EU AI Act provision excerpt speaks to US law or executive orders; un-adjudicable from this corpus. | Claim about the US lacking a federal counterpart, the January 2025 executive order rescinding safeguards, and the December 2025 'national policy framework' order. Sourced t","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Operative mechanics: a risk-tiered, product-safety architecture","sentence":"Providers of such systems carry the heaviest burden: a risk-management system (Art. 9), data-governance duties (Art. 10), technical documentation (Art. 11), logging (Art. 12), transparency and instructions for use (Art. 13), human oversight by design (Art. 14), and accuracy/robustness/cybersecurity (Art. 15), all funnelled through ex-ante conformity assessment and CE-marking (Arts. 16, 43).","judgeVerdict":"out_of_corpus","rationale":"Lists provider obligations under Arts. 9-16/43; no excerpt covers these provider articles (the Art. 26 excerpt concerns deployers).","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Cross-jurisdiction position: the only binding horizontal regime","sentence":"(EU) 2024/1689 applies via Art. 2 to providers placing systems on the EU market regardless of establishment).","judgeVerdict":"out_of_corpus","rationale":"Asserts extraterritorial reach under Art. 2 to providers regardless of establishment; the only Art. 2 excerpt is the military/defence exclusion, a different aspect that neither establishes nor contradicts this.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Operative mechanics: a risk-tiered, product-safety architecture","sentence":"At the apex, Art. 5(1) prohibits a closed list of practices deemed to pose unacceptable risk — including social scoring (Art. 5(1)(c)), untargeted scraping of facial images to build recognition databases (Art. 5(1)(e)), emotion inference in workplaces and education (Art. 5(1)(f)), and (subject to narrow law-enforcement carve-outs) real-time remote biometric identification in publicly accessible spaces (Art. 5(1)(h)).","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: the Art. 5 excerpt covers only real-time remote biometric identification for law enforcement. Social scoring (5(1)(c)), untargeted facial scraping (5(1)(e)), and workplace/education emotion inference (5(1)(f)) are distinct enumerated factual limbs ","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Operative mechanics: a risk-tiered, product-safety architecture","sentence":"Enforcement bites through Art. 99's tiered administrative fines: up to EUR 35 million or 7% of worldwide annual turnover for prohibited-practice breaches, EUR 15 million / 3% for most other operator obligations, and EUR 7.5 million / 1% for supplying incorrect information.","judgeVerdict":"out_of_corpus","rationale":"Asserts specific Art. 99 fine figures (EUR 35M/7% etc.); no excerpt covers penalties or fine amounts.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Implementation and trajectory: staggered timeline and the Digital Omnibus","sentence":"The Act entered into force on 1 August 2024 (twenty days after Official Journal publication of Reg.","judgeVerdict":"out_of_corpus","rationale":"Asserts an entry-into-force date (1 August 2024); no excerpt establishes dates.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Cross-jurisdiction position: the only binding horizontal regime","sentence":"State experimentation has likewise pivoted away from the EU template — Colorado's 2024 AI Act, the closest US analogue with its developer/deployer split and impact-assessment duties, was reworked in 2026 toward a transparency-and-recordkeeping model rather than EU-style conformity assessment (Mayer Brown 2026).","judgeVerdict":"out_of_corpus","rationale":"Describes Colorado's 2024 AI Act, a separate instrument not in the excerpts.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Cross-jurisdiction position: the only binding horizontal regime","sentence":"China's regime is binding but vertical and rolled out piecemeal — the Interim Measures for the Management of Generative AI Services (effective 15 August 2023) target public-facing generative services with content-control and security-assessment duties, and scholarship reads them partly as a pro-growth signalling device rather than a comprehensive risk framework (Zhang, 'The Promise and Perils of China's Regulation of Artificial Intelligence,' Columbia J.","judgeVerdict":"out_of_corpus","rationale":"Describes China's Interim Measures for Generative AI Services, a separate instrument not in the excerpts.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Cross-jurisdiction position: the only binding horizontal regime","sentence":"Comparative scholarship situates these as three distinct 'rulebooks' — rights-based (EU), state-control (China), and market-driven (US) — competing for influence (Smuha 2021, doi:10.1080/17579961.2021.1898300), with the Act's compute-threshold approach to GPAI (Art. 51(2)) now the most-emulated technical mechanism abroad.","judgeVerdict":"out_of_corpus","rationale":"A comparative 'three rulebooks' framing and claim that the compute threshold is most-emulated abroad; the sentence's assertion is about emulation/comparison, which no excerpt establishes.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Implementation and trajectory: staggered timeline and the Digital Omnibus","sentence":"Implementation infrastructure followed the timeline unevenly: the European Commission's AI Office was established to supervise GPAI, and a voluntary GPAI Code of Practice (published by the AI Office on 10 July 2025) with accompanying Commission scope guidelines (published 18 July 2025) was issued to bridge the gap until harmonised standards exist — guidance that had to stabilise still-contested boundaries between the Act's 'AI system', 'general-purpose AI model', and 'foundation model' categories (Fernández-Llorca et al. 2025, doi:10.1007/s10506-024-09412-y).","judgeVerdict":"out_of_corpus","rationale":"Concerns the AI Office, GPAI Code of Practice, and contested category boundaries; no excerpt covers implementation infrastructure or these guidance documents.","inReviewQueue":false},{"instrument":"EU-AIA-2024","slug":"eu-ai-act","section":"Implementation and trajectory: staggered timeline and the Digital Omnibus","sentence":"The net trajectory is one of phased, standards-contingent application in which the EU pursues AI governance through interlocking law and policy (Hulok 2025, doi:10.1007/s12027-025-00869-1): the rights-protective core remains binding, but the operative high-risk machinery now hinges on whether the deferred harmonised-standards and tooling milestones are met before the revised 2027-2028 dates.","judgeVerdict":"out_of_corpus","rationale":"Synthesis about phased, standards-contingent application and revised 2027-2028 dates; no excerpt establishes standards dependency or timeline milestones.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Operative mechanics: strict liability re-engineered for software and AI","sentence":"The redress machinery is the load-bearing element: Art. 6 fixes compensable damage (death, personal injury including medically recognised psychological harm, property, and destruction or corruption of non-professional data); Art. 8 names the chain of liable economic operators down to importers, authorised representatives, fulfilment-service providers and certain platforms; Art. 9 compels disclosure of evidence in a defendant's control; and Art. 10 supplies rebuttable presumptions of defect and causation.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: only Art. 10's rebuttable presumptions are in corpus. Art. 6 (compensable-damage categories), Art. 8 (chain of liable operators), and Art. 9 (evidence disclosure) are three distinct factual limbs entirely absent from the single Art. 10 excerpt. Ove","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"The complexity presumption and the opacity problem","sentence":"Scholarship on visibility and identification of AI systems (arXiv:2406.12137; 10.1145/3630106.3658948, Chan et al. 2024) underscores why such infrastructural evidence matters — without IDs, logs and monitoring, even a rebuttable presumption struggles for factual purchase.","judgeVerdict":"out_of_corpus","rationale":"The claim concerns scholarship on AI-system visibility/identification (arXiv and ACM citations); no excerpt speaks to this.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Operative mechanics: strict liability re-engineered for software and AI","sentence":"Art. 4(1) brings \"software\" — including standalone AI systems — within the meaning of \"product,\" and Recital 13 treats an AI system provider within the meaning of Regulation (EU) 2024/1689 as a manufacturer irrespective of delivery model.","judgeVerdict":"out_of_corpus","rationale":"The claim about Art. 4(1) treating software/AI as a 'product' and Recital 13's manufacturer treatment is not addressed by the sole Art. 10 presumption excerpt.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"The complexity presumption and the opacity problem","sentence":"This is reinforced by the Art. 9 disclosure duty and the Art. 10(2)(a) adverse inference when a defendant withholds ordered evidence, so opacity itself becomes a litigation cost for producers rather than for victims.","judgeVerdict":"out_of_corpus","rationale":"The claim concerns the Art. 9 disclosure duty and the Art. 10(2)(a) adverse inference for withholding evidence, which the excerpt (excessive-difficulty presumption) does not cover.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Agentic systems and post-market mutability","sentence":"Art. 7(2)(c) makes a product's \"ability to continue to learn or acquire new features after it is placed on the market\" relevant to assessing defectiveness, abandoning the assumption that a product is frozen at the moment of sale.","judgeVerdict":"out_of_corpus","rationale":"The claim about Art. 7(2)(c) treating post-market learning as relevant to defectiveness has no on-point excerpt.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"The complexity presumption and the opacity problem","sentence":"The mechanism interlocks with Regulation (EU) 2024/1689: a proven breach of AI Act logging, transparency or risk-management obligations can feed the presumptions.","judgeVerdict":"out_of_corpus","rationale":"The claim that a proven AI Act breach can feed the presumptions is not established by the excerpt, which addresses only the excessive-difficulty presumption without reference to the AI Act.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Fault lines, status, and implementation trajectory","sentence":"Contestability research (10.1145/3757415, Yurrita et al. 2025; Karusala et al. 2024) warns that monetary liability is a thin substitute for meaningful, accessible avenues to challenge automated decisions.","judgeVerdict":"out_of_corpus","rationale":"The claim cites contestability research about monetary liability being a thin substitute for redress; no excerpt is on-point.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"The complexity presumption and the opacity problem","sentence":"Art. 10(4) instructs a national court to presume defectiveness or the causal link where a claimant faces \"excessive difficulties, in particular due to technical or scientific complexity,\" in proving them — a direct response to the black-box character of machine-learning systems whose internal logic claimants cannot reconstruct.","judgeVerdict":"supported","rationale":"The claim quotes Art. 10(4)'s presumption of defectiveness or causal link where a claimant faces 'excessive difficulties, in particular due to technical or scientific complexity,' matching the excerpt directly.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Agentic systems and post-market mutability","sentence":"Art. 11(2) keeps manufacturers liable for defects introduced by software updates or upgrades, and by machine learning, that remain within their control — narrowing the traditional later-defect defence.","judgeVerdict":"out_of_corpus","rationale":"The claim about Art. 11(2) keeping manufacturers liable for software/ML update defects within their control has no on-point excerpt.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Fault lines, status, and implementation trajectory","sentence":"The Directive is adopted but not yet in force: it entered into force on 18 November 2024, yet its substantive liability rules apply only to products placed on the market after 9 December 2026 (Art. 2(1)), leaving a transposition and adaptation window for Member States and producers.","judgeVerdict":"out_of_corpus","rationale":"The claim about entry-into-force and application dates (18 Nov 2024 / 9 Dec 2026, Art. 2(1)) is not represented among the excerpts.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Agentic systems and post-market mutability","sentence":"These rules speak to the governance gap around autonomous and continually-adapting agents, a frontier that legal scholarship treats as distinct from static models: Kolt's agency-law framing (arXiv:2501.07913) and the agent-infrastructure and delegation proposals (arXiv:2501.10114, Chan et al. 2025; arXiv:2501.09674) argue that attribution and remediation require external systems, while multi-agent failure modes (arXiv:2502.14143) complicate the \"within the manufacturer's control\" line that Art. 11(2) draws.","judgeVerdict":"out_of_corpus","rationale":"The claim concerns agent-governance scholarship and Art. 11(2)'s 'within the manufacturer's control' line; no excerpt is on-point.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Fault lines, status, and implementation trajectory","sentence":"Its salience grew when the Commission withdrew the separate proposed AI Liability Directive in 2025, so the PLD now carries the principal EU AI-liability load while remaining deliberately silent on ex-ante topics — biometrics, deepfakes and compute fall to Regulation (EU) 2024/1689 and sectoral law.","judgeVerdict":"out_of_corpus","rationale":"The claim about the withdrawal of the proposed AI Liability Directive and the PLD's silence on ex-ante topics has no supporting excerpt.","inReviewQueue":false},{"instrument":"EU-PLD-2024","slug":"eu-product-liability-directive","section":"Operative mechanics: strict liability re-engineered for software and AI","sentence":"The revised Directive (EU) 2024/2853 keeps the 1985 regime's no-fault core but redefines its perimeter for the digital economy.","judgeVerdict":"out_of_corpus","rationale":"The claim that the revised Directive keeps the 1985 no-fault core while redefining its digital perimeter is not addressed by the sole Art. 10 excerpt.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Key Fault-Lines and Critiques","sentence":"A second gap is scope: the Art. 7 data-processing prohibitions are tightly drawn, leaving most performance-scoring and task-allocation analytics permissible so long as they avoid the enumerated categories — a critique that echoes Adams-Prassl and co-authors' warning that piecemeal rules leave the core ranking-and-deactivation machinery of algorithmic management largely untouched (10.1177/20319525231167299).","judgeVerdict":"out_of_corpus","rationale":"Excerpt 3 confirms Art. 7 bars biometric one-to-many identity processing, but nothing in the corpus establishes that the prohibitions are 'tightly drawn' leaving performance-scoring permissible, nor the Adams-Prassl critique.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Implementation Trajectory","sentence":"Whether the Directive's transparency (Art. 9) and human-oversight (Art. 10) duties translate into genuine worker agency, or merely procedural compliance, is the open question its first transposition cycle will test.","judgeVerdict":"supported","rationale":"Excerpt 4 establishes Art. 9 informs workers about automated monitoring (transparency) and Excerpt 1 establishes Art. 10 requires human oversight, matching the claim's article-to-duty mapping.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Key Fault-Lines and Critiques","sentence":"The right to explanation the Directive gestures at inherits this limitation: meaningful contestation may require counterfactual explanations of the kind Wachter and colleagues propose (arXiv:1711.00399), which the text does not mandate in technical form.","judgeVerdict":"out_of_corpus","rationale":"No excerpt addresses counterfactual explanations, the Wachter proposal, or whether the text mandates them in technical form.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Cross-Jurisdiction Position","sentence":"Unlike the horizontal Regulation (EU) 2024/1689 (the AI Act), which classifies workplace AI as high-risk but regulates the system, the Directive regulates the employment relationship itself, so the two instruments overlap rather than coincide.","judgeVerdict":"out_of_corpus","rationale":"No excerpt discusses the AI Act (Reg. 2024/1689) or the contrast that the Directive 'regulates the employment relationship itself.'","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Key Fault-Lines and Critiques","sentence":"Transposition discretion before December 2026 is a further source of fragmentation risk.","judgeVerdict":"out_of_corpus","rationale":"No excerpt addresses transposition discretion, the December 2026 date, or fragmentation risk.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Key Fault-Lines and Critiques","sentence":"A statutory right to human review (Art. 11) presumes the reviewer can meaningfully interrogate the model, yet Bayamlioglu shows that contestation rights under the GDPR have foundered precisely because data subjects lack the information and counterfactual reasoning needed to mount a challenge (10.1111/rego.12391).","judgeVerdict":"out_of_corpus","rationale":"Excerpt 2 confirms Art. 11 gives a right to human review, but the sentence's actual assertion — the Bayamlioglu/GDPR contestation-failure critique — is not represented in any excerpt.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Implementation Trajectory","sentence":"The stakes are framed by the broader trajectory of automation in labour markets: Acemoglu and Restrepo model how algorithmic technologies both displace and reinstate tasks, with distributional effects that hinge on whether institutions channel them (10.1257/jep.33.2.3), while the earlier susceptibility estimates of Frey and Osborne (10.1016/j.techfore.2016.08.019) catalysed the policy attention that instruments like this Directive now answer.","judgeVerdict":"out_of_corpus","rationale":"No excerpt addresses Acemoglu-Restrepo, Frey-Osborne, or labour-market automation dynamics.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Operative Mechanics: Two Pillars over Platform Work","sentence":"Directive (EU) 2024/2831, in force since 1 December 2024 with a transposition deadline of 2 December 2026, rests on two pillars.","judgeVerdict":"out_of_corpus","rationale":"No excerpt states the directive number 2024/2831, the 1 December 2024 in-force date, the transposition deadline, or the 'two pillars' framing.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Cross-Jurisdiction Position","sentence":"Where Wachter, Mittelstadt and Floridi argued that the GDPR's Article 22 confers at most a limited 'right to be informed' rather than a genuine right to explanation of automated decisions (10.1093/idpl/ipx005), Art. 9 and Art. 11 of the Platform Work Directive impose concrete, sector-specific information and human-review duties on platforms, including a right to a written statement of reasons for significant decisions.","judgeVerdict":"supported","rationale":"Excerpt 4 establishes Art. 9 imposes information duties and Excerpt 2 establishes Art. 11 provides a written explanation plus human review of significant decisions, matching the claim's concrete Art. 9/Art. 11 duties.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Implementation Trajectory","sentence":"Member States must transpose the Directive by 2 December 2026, and its substantive bite will depend on national choices about the presumption's triggers and the resourcing of labour inspectorates.","judgeVerdict":"out_of_corpus","rationale":"No excerpt addresses the 2 December 2026 transposition deadline, presumption triggers, or labour inspectorate resourcing.","inReviewQueue":false},{"instrument":"EU-PWD-2024","slug":"eu-platform-work-directive","section":"Operative Mechanics: Two Pillars over Platform Work","sentence":"Art. 7 prohibits processing of data on a worker's emotional or psychological state, private communications, biometric data used to establish identity by one-to-many comparison, and inferences about protected characteristics or trade-union activity; Art. 8 mandates a data protection impact assessment; Art. 9 requires transparency to workers and their representatives about automated monitoring and decision-making systems; Art. 10 requires human oversight by competent staff who can override the system; and Art. 11 requires human review of, and reasons for, significant decisions, barring account restriction, suspension or termination by solely automated means.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Heavily multi-limb: the Art. 7 excerpt covers only the biometric one-to-many identity prohibition, not emotional/psychological-state, private-communications, or protected-characteristic/trade-union inference limbs; Art. 8 (DPIA) is absent entirely. Arts. 9-11 ","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"The Prioritization Experiment: What Concretely Changed","sentence":"The governance gate moved too: OMB M-25-21 rescinded and replaced M-24-10, retaining Chief AI Officers and use-case inventories while narrowing minimum risk practices to 'high-impact' AI (OMB M-25-21, Apr. 3, 2025).","judgeVerdict":"out_of_corpus","rationale":"About OMB M-25-21 rescinding/replacing M-24-10 and narrowing minimum risk practices to 'high-impact' AI. The sole committed excerpt covers SSP/NIST SP 800-53 documentation only; the OMB governance-memo claim is un-excerpted. | Claim about OMB M-25-21 rescinding and replacing M-24-10, retaining Chief AI Officers/use-case inventories and narrowing minimum risk practices to 'high-impact' AI. The only","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"Adoption Trajectory and Outlook","sentence":"This mirrors EU security-carveout dynamics, where Palmiotto traces widening law-enforcement exceptions producing double standards (10.1017/err.2024.97), Yazici flags under-regulated biometric and satellite surveillance (10.1080/17579961.2025.2470589), and Statewatch warns exemptions make oversight 'extremely difficult' (Jones and Lanneau 2025).","judgeVerdict":"out_of_corpus","rationale":"The single excerpt concerns FedRAMP SSP/NIST 800-53 disclosure and says nothing about EU security carveouts, law-enforcement exceptions, or the Palmiotto/Yazici/Statewatch dynamics this claim asserts.","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"What the Guidance Commits To","sentence":"Its core commitment is integrative rather than novel: AI and generative-AI cloud services that process federal data must obtain a FedRAMP Authorization to Operate at the Low, Moderate, or High baseline, on the same authorisation rails as any other cloud service — a posture that matches the breadth of LLMs as a general-purpose technology, which Eloundou et al. estimate could affect at least 10% of work tasks for roughly 80% of the U.S. workforce (10.1126/science.adj0998).","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] The excerpt supports that FedRAMP authorization requires an SSP documenting 800-53 controls, but the load-bearing specific claim that AI services must obtain an ATO 'at the Low, Moderate, or High baseline' is a distinct factual detail absent from the excerpt. ","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"The Prioritization Experiment: What Concretely Changed","sentence":"Crucially, no AI control baseline was created: prioritization sat 'on top of existing FedRAMP Authorization paths' and moved vendors near the front of the queue - 'the authorization itself will take a similar amount of time' (FedRAMP Draft ET Prioritization Framework, Jan. 26, 2024).","judgeVerdict":"out_of_corpus","rationale":"About the Jan-2024 draft prioritization framework sitting atop existing FedRAMP paths without creating an AI baseline, with a verbatim quote. The single SSP-documentation excerpt does not address the prioritization framework; un-adjudicable. | Claim that no AI control baseline was created and prioritization sat atop existing FedRAMP paths with unchanged authorization time, quoting the Jan 2024 dra","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"Standing Relative to Binding Law","sentence":"Critically, FedRAMP authorisation enables but does not by itself approve a specific AI use case — the agency authorising official remains the operative gate, and OMB governance applies separately (originally via M-24-10, which M-25-21 rescinded and replaced in April 2025 - see the prioritization record below).","judgeVerdict":"out_of_corpus","rationale":"About the agency authorising official as the operative gate and OMB M-24-10/M-25-21 governance applying separately. The sole excerpt (SSP/NIST controls) does not establish authorizing-official gating or the OMB memo history; un-excerpted. | Claim that FedRAMP authorisation enables but does not approve a specific AI use case, that the agency authorising official is the operative gate, and that OMB ","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"The Prioritization Experiment: What Concretely Changed","sentence":"The final criteria gated the fast lane on demand, not security: a demand score of at least 3, with current federal customers worth 1 point each (minimum one), indirect 0.5, and potential 0.25 (FedRAMP Emerging Technologies Prioritization Criteria and Guidance V3, June 2024).","judgeVerdict":"out_of_corpus","rationale":"Specific demand-score criteria (>=3, customers 1/0.5/0.25 points) from the June-2024 prioritization guidance. The single SSP-documentation excerpt does not cover any prioritization criteria; un-adjudicable figures. | Detailed claim about the final demand-score criteria (score >=3; customer point weights). The only corpus excerpt is the SSP/control documentation summary, which contains no scoring/p","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"Critiques and Structural Gaps","sentence":"Supply-chain risk-management for model and dataset provenance is gestured at in the SSP but lacks granular disclosure machinery — Ruschemeier shows foundation models can memorise and leak training data, defeating anonymity assumptions (10.1017/cfl.2024.2), and Havlikova shows provenance opt-outs fail post-LAION (JIPITEC, view/422).","judgeVerdict":"out_of_corpus","rationale":"[adversarial re-audit correction] The excerpt establishes that provenance disclosure IS extended (training-data provenance, evaluation results, model documentation) but does NOT establish the sentence’s operative defect-claim that it ‘lacks granular disclosure machinery’ — an adequacy judgment absent from the corpus. Downgraded supported→out_of_corpus.","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"What the Guidance Commits To","sentence":"Agencies and the authorisation pathway are directed to weigh model-specific risks — training-data exposure, prompt-injection, and unintended output disclosure — when scoping the System Security Plan and selecting NIST SP 800-53 control overlays; Ruschemeier shows why this matters, since models that 'memorize and leak pieces of training data' defeat ordinary anonymity assumptions (10.1017/cfl.2024.2).","judgeVerdict":"uncertain","rationale":"The excerpt covers disclosure of provenance/evaluation/model documentation, but does not direct agencies to weigh prompt-injection, unintended-output-disclosure risks, or select NIST 800-53 control overlays, so that specific assertion is not represented.","inReviewQueue":true},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"What the Guidance Commits To","sentence":"FedRAMP's 2024 AI cloud guidance is operational PMO direction issued under the program's standing statutory base (44 U.S.C. §3607), not a freestanding rule.","judgeVerdict":"out_of_corpus","rationale":"The excerpt says nothing about the guidance being operational PMO direction, its statutory base 44 U.S.C. §3607, or its status as not a freestanding rule.","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"The Prioritization Experiment: What Concretely Changed","sentence":"The January 26, 2024 draft was unusually granular: eligible offerings had to run on a 'foundation model' with 'at least tens of billions of parameters,' make generative AI the primary purpose (capabilities 'embedded within a broader product' might not qualify), and cite at least one third-party benchmark from a named menu (WinoGrande, ARC-Challenge, HellaSwag, OpenBookQA, MMLU 5-shot, HumanEval, and MBPP for chat; HumanEval/MBPP for code; CLIPScore and X-IQE-Overall for images), disclosing any benchmark-developer affiliation (FedRAMP Draft Emerging Technology Prioritization Framework, Jan. 26, 2024).","judgeVerdict":"out_of_corpus","rationale":"Granular Jan-2024 draft eligibility details (tens-of-billions parameters, benchmark menu WinoGrande/MMLU/HumanEval/etc.). The single SSP-documentation excerpt covers none of these; entirely un-excerpted specifics. | Highly specific claim about the Jan 26 2024 draft's eligibility (foundation model with tens of billions of parameters, generative AI as primary purpose, named benchmark menu, developer","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"What the Guidance Commits To","sentence":"The guidance also cross-walks to OMB M-24-10 minimum practices for safety- and rights-impacting AI, positioning FedRAMP as the security gate beneath a separate governance gate - a cross-walk that now points at a rescinded target, since OMB M-25-21 rescinded and replaced M-24-10 in April 2025 (see the prioritization record below).","judgeVerdict":"out_of_corpus","rationale":"About the guidance cross-walking to OMB M-24-10 minimum practices now rescinded by M-25-21. The single SSP-documentation excerpt does not mention any OMB cross-walk; un-excerpted. | Claim about the guidance cross-walking to OMB M-24-10 minimum practices, positioning FedRAMP as a security gate, and that cross-walk now pointing at a rescinded target (M-25-21). The corpus excerpt only summarizes SSP/","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"The Prioritization Experiment: What Concretely Changed","sentence":"That thinness cut both ways: it was trivially rescindable - eliminated when Executive Order 14148 (January 20, 2025) revoked EO 14110, with Executive Order 14179 (January 23, 2025) casting the prior regime as 'onerous and unnecessary government control,' already-authorized vendors untouched since only queue order was at stake (Winvale 2025) - and the substantive SP 800-53 tailoring is now being built outside FedRAMP: NIST's Control Overlays for Securing AI Systems (COSAiS) concept paper went out for comment August 14, 2025, covering generative AI, predictive AI, single- and multi-agent systems, and AI developers (first discussion draft Jan. 8, 2026) (NIST 2026).","judgeVerdict":"out_of_corpus","rationale":"The lone corpus excerpt only establishes that FedRAMP authorisation requires an SSP documenting NIST SP 800-53 controls and that GenAI guidance extends disclosure to training-data provenance/evaluations/model docs. It says nothing on-point about any of the sentence's actual load-bearing claims: EO 1","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"Standing Relative to Binding Law","sentence":"Its legal force derives from FedRAMP's authorisation mandate, not from new rulemaking, so it tailors existing obligations rather than creating fresh ones.","judgeVerdict":"supported","rationale":"The excerpt describes GenAI guidance as extending existing FedRAMP SSP/NIST 800-53 disclosure obligations, on-point to and establishing the claim that the guidance tailors existing obligations rather than creating fresh ones.","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"Adoption Trajectory and Outlook","sentence":"So far, though, that maturation is advancing outside FedRAMP: as the prioritization record below shows, the substantive SP 800-53 tailoring has migrated to NIST's COSAiS overlays, and the program's one concrete GenAI instrument - demand-keyed prioritization - was rescinded in January 2025.","judgeVerdict":"out_of_corpus","rationale":"About SP 800-53 tailoring migrating to NIST COSAiS and prioritization being rescinded Jan-2025. Neither COSAiS migration nor the rescission is in the single SSP-documentation excerpt; un-adjudicable. | Claim that substantive SP 800-53 tailoring migrated to NIST's COSAiS overlays and that demand-keyed prioritization was rescinded in January 2025. The sole corpus excerpt describes what FedRAMP autho","inReviewQueue":false},{"instrument":"FEDRAMP-AI-2024","slug":"fedramp-ai-guidance","section":"The Prioritization Experiment: What Concretely Changed","sentence":"Executive Order 14110 Sec. 10.1(f)(ii) gave GSA 90 days to issue a FedRAMP prioritization framework 'starting with generative AI offerings' - LLM chat interfaces, code-generation tools, and prompt-based image generators - to apply for no less than 2 years (88 Fed.","judgeVerdict":"out_of_corpus","rationale":"About EO 14110 Sec. 10.1(f)(ii) directing GSA to issue a prioritization framework. The single SSP/NIST-controls excerpt does not cover the EO mandate or the 90-day/2-year terms; un-excerpted external mandate. | Claim about EO 14110 Sec. 10.1(f)(ii) giving GSA 90 days to issue a prioritization framework starting with generative AI, applying for at least 2 years, with a Federal Register cite. The si","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"Critiques and Structural Gaps","sentence":"Disclosure obligations presume providers can characterize corpora, yet foundation models 'memorize and leak pieces of training data' and so resist treatment as anonymous (Ruschemeier 2025, 10.1017/cfl.2024.2), while opt-out and rights-clearance mechanisms remain practically brittle post-LAION (Havlikova 2025) and contested across copyright regimes (Kretschmer et al. 2025, 10.1093/grurint/ikaf093) — terrain on which the research-TDM route is even proposed as a conditional 'safe harbor' for openly released models (Radeisen 2026, 10.1093/grurint/ikag002), underscoring how unsettled provenance obligations remain.","judgeVerdict":"out_of_corpus","rationale":"The claim concerns foundation models memorizing/leaking training data and copyright/opt-out regimes via scholarly citations; no excerpt is on-point to memorization, leakage, or copyright provenance debates.","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"Adoption Trajectory and Outlook","sentence":"Comparative pressure may push it toward firmer disclosure: EU debates over general-purpose and foundation-model documentation, and security-exemption critiques warning that meaningful oversight of policing and security AI is becoming 'extremely difficult' (Jones & Lanneau 2025; Palmiotto 2025, 10.1017/err.2024.97), foreshadow where US procurement terms on provenance, evaluation, and national-security carveouts (FAR Subpart 4.21) will face tightening or contestation — pressure echoed in findings that defence and national-security exclusions leave biometric and surveillance uses under-regulated (Yazici 2025, 10.1080/17579961.2025.2470589).","judgeVerdict":"out_of_corpus","rationale":"This is about EU foundation-model documentation debates, security-exemption critiques, FAR Subpart 4.21, and national-security carveouts — none of which appear in the excerpts.","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Primary Text Actually Contains","sentence":"The introduction labels it \"version 1.0\", offers \"prompts to consider and frame your thinking\" rather than \"directive recommendations\", and states: \"This content is non-binding\" (GSA GenAI Guide v1.0 Introduction, Apr. 29, 2024).","judgeVerdict":"out_of_corpus","rationale":"Quotes the guide's 'version 1.0', 'prompts to consider', and 'This content is non-binding' framing. None of the three faithful-summary excerpts (discrete category, SINs, vendor disclosure) speaks to the guide's version label or binding status; un-adjudicable. | Claim quoting the guide's introduction: 'version 1.0', 'prompts to consider', 'directive recommendations', 'This content is non-binding.' ","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"Critiques and Structural Gaps","sentence":"Most consequentially, redress is only implicit — pushed onto M-24-10 — leaving acquisition language thin on what makes contestation meaningful for decision subjects (Yurrita et al. 2025, 10.1145/3757415; Schmude et al. 2025, arXiv:2504.18236).","judgeVerdict":"out_of_corpus","rationale":"The assertion that redress is pushed onto M-24-10 and that contestation is thin is not addressed by any excerpt.","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Primary Text Actually Contains","sentence":"It also maps agency-specific and non-FAR paths: Army CHESS, DHS EAGLE Next Gen, phased \"Pilot IRS\", DoD Tradewinds, CRADAs, Economy Act agreements, OTAs, and Challenge.gov prizes (GSA Guide v1.0 Sec. 4, 2024).","judgeVerdict":"out_of_corpus","rationale":"Enumerates specific agency/non-FAR paths (Army CHESS, DHS EAGLE, DoD Tradewinds, CRADAs, OTAs, Challenge.gov). None of these appear in the corpus; the SINs excerpt is about MAS SINs, not these vehicles, so the specific list is un-excerpted. | Claim enumerating agency-specific and non-FAR acquisition paths (Army CHESS, DHS EAGLE Next Gen, Pilot IRS, DoD Tradewinds, CRADAs, Economy Act, OTAs, Challe","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Primary Text Actually Contains","sentence":"Section 4 concedes \"There is currently not a governmentwide Generative AI-only acquisition vehicle\", routing buyers to whole vehicles - MAS IT (formerly Schedule 70, 7.5 million-plus offerings) and Best-in-Class GWACs (8(a) STARS III, Alliant 2, CIO-SP3, EIS, NASA SEWP, VETS 2) - and enumerates no Special Item Numbers; 54151S appears nowhere.","judgeVerdict":"out_of_corpus","rationale":"The corpus 'Acquisition-vehicle routing' excerpt establishes the routing to MAS IT and Best-in-Class GWACs, the absence of a generative-AI-only vehicle, and that no new AI-specific SINs are enumerated. However, the sentence piles on multiple specifics NOT in any excerpt: the verbatim quotation ('The","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"Standing Relative to Binding Law","sentence":"This indirection mirrors a broader law-and-policy hybrid pattern in AI governance, where soft guidance scaffolds harder obligations (Hulok 2025, 10.1007/s12027-025-00869-1).","judgeVerdict":"out_of_corpus","rationale":"The claim about soft guidance scaffolding harder obligations is a scholarly framing not represented in the excerpts.","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Primary Text Actually Contains","sentence":"GSA released the guide April 29, 2024, a 180-day EO 14110 Section 10.1(h) deliverable, not in September (GSA news release, Apr. 29, 2024).","judgeVerdict":"out_of_corpus","rationale":"About the April-29-2024 release date and EO 14110 Section 10.1(h) 180-day deliverable status. No committed excerpt establishes the issuance date or the EO-deliverable framing; date/external-mandate claim un-adjudicable. | Claim about the release date (April 29 2024) as a 180-day EO 14110 Section 10.1(h) deliverable, correcting a 'September' error. This is a date/provenance claim; the three corpus ","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Guide Directs Agencies to Do","sentence":"By embedding responsible-AI considerations into requirements derivation rather than statute, it helps agencies translate OMB M-24-10's minimum-practice expectations into contractible obligations — though it leans on category terms like 'foundation model' that remain definitionally unstable even where they are legislated directly, as the EU AI Act's shifting text shows (Fernández-Llorca et al. 2025, 10.1007/s10506-024-09412-y).","judgeVerdict":"out_of_corpus","rationale":"Multi-limb claim about embedding responsible-AI into requirements derivation, translating OMB M-24-10 practices, 'foundation model' definitional instability, and the EU AI Act, with a scholarly citation. The corpus does not cover OMB M-24-10, definitional instability, or the EU Act; the on-point core is only partially covered. | Claim that embedding responsible-AI into requirements derivation help","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Primary Text Actually Contains","sentence":"That layer proved as policy-contingent as feared: M-25-22 rescinded and replaced M-24-18 under EO 14179, pivoting to vendor lock-in protections and pre-award testing of \"high-impact AI\" from September 30, 2025 (OMB M-25-22, Apr. 3, 2025).","judgeVerdict":"out_of_corpus","rationale":"About OMB M-25-22 rescinding/replacing M-24-18 under EO 14179 and pivoting to lock-in protections/high-impact-AI testing. None of the three faithful-summary excerpts covers these OMB memos or EO; un-excerpted external policy. | Claim about M-25-22 rescinding and replacing M-24-18 under EO 14179 and pivoting to vendor lock-in protections and pre-award testing of high-impact AI from Sept 30 2025. Th","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"Standing Relative to Binding Law","sentence":"The salience of that translation is amplified by how broadly these systems reach: roughly 80% of the U.S. workforce \"could have at least 10% of their work tasks affected\" by LLMs, which display \"traits of general-purpose technologies\" (Eloundou et al. 2024, 10.1126/science.adj0998).","judgeVerdict":"out_of_corpus","rationale":"The 80%-of-workforce / general-purpose-technology figures are a scholarly citation (Eloundou et al.) with no on-point excerpt.","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"Adoption Trajectory and Outlook","sentence":"Because the Guide is operational rather than legislative, uptake hinges on agencies actually importing its due-diligence questions and suggested provisions into solicitation packages — the notes indicate agencies typically do, though the primary text itself ships no sample clauses and enumerates no SINs (see the primary-text section above), so any standardization channel, even absent a mandate, runs through the existing vehicles it maps rather than a dedicated SIN structure.","judgeVerdict":"out_of_corpus","rationale":"The no-SIN and existing-vehicle-routing limbs are corpus-supported, but the sentence's load-bearing empirical claim — 'the notes indicate agencies typically do [import the due-diligence questions/provisions into solicitation packages]' — is an adoption/uptake assertion absent from every excerpt (the","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Primary Text Actually Contains","sentence":"The binding vendor-disclosure regime - agencies \"must consider\" requiring sub-group performance metrics and training-data \"source, provenance, selection, quality\" - arrived separately in OMB M-24-18, covering solicitations 180-plus days out and exempting National Security Systems (OMB M-24-18, Sept. 24, 2024).","judgeVerdict":"out_of_corpus","rationale":"About the binding vendor-disclosure regime arriving separately in OMB M-24-18 (180-plus-day solicitations, NSS exemption). The vendor-disclosure excerpt describes the guide's own sample language, not OMB M-24-18; the memo, dates, and NSS carve-out are un-excerpted. | Claim about the binding vendor-disclosure regime arriving separately in OMB M-24-18 (sub-group performance metrics; training-data so","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Guide Directs Agencies to Do","sentence":"Issued April 29, 2024 (GSA, Generative AI and Specialized Computing Infrastructure Acquisition Resource Guide), the Guide operates through procurement plumbing rather than command.","judgeVerdict":"out_of_corpus","rationale":"Core is the April-29-2024 issuance date plus the full title, characterizing the guide as operating 'through procurement plumbing rather than command.' No committed excerpt establishes the issuance date or the binding-versus-operational character; un-adjudicable. | Claim about the issuance date (April 29 2024) plus the guide's full title, framing it as operating 'through procurement plumbing rather","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Guide Directs Agencies to Do","sentence":"It routes generative-AI and foundation-model buys as a discrete acquisition category, mapping them onto existing governmentwide vehicles — MAS IT and the Best-in-Class GWACs — so agencies can channel and surface AI spend through established channels, though the primary text itself enumerates no Special Item Numbers (see the primary-text section below).","judgeVerdict":"supported","rationale":"All core assertions have on-point excerpts: 'Generative AI acquisition guidance' establishes that GenAI/foundation-model acquisition is treated as a discrete category; 'Acquisition-vehicle routing' establishes mapping onto MAS IT and Best-in-Class GWACs and that no AI-specific SINs are enumerated. '","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Primary Text Actually Contains","sentence":"Its hardest edges are supply-chain cross-references: Section 889 at FAR 4.2102(a)(2), FY23 NDAA Section 5949 semiconductor prohibitions effective December 2027, the TAA clause at FAR 52.225-5, and DFARS 239.7602-2 / HSAR 3052.204-72 data-location rules (GSA Guide v1.0 Sec. 5.4, 2024).","judgeVerdict":"out_of_corpus","rationale":"Supply-chain cross-references (Section 889 at FAR 4.2102(a)(2), NDAA Section 5949, TAA at FAR 52.225-5, DFARS 239.7602-2 / HSAR 3052.204-72). None of the three faithful-summary excerpts covers these clauses; entirely un-excerpted specifics. | Claim listing supply-chain cross-references (FAR 4.2102(a)(2) Section 889, FY23 NDAA Section 5949 semiconductor prohibitions effective Dec 2027, TAA clause F","inReviewQueue":false},{"instrument":"GSA-AI-GUIDE-2024","slug":"gsa-ai-acquisition-guide","section":"What the Primary Text Actually Contains","sentence":"Nor does it ship sample clauses: Section 3.4 offers supplier due-diligence questions instead - data provenance and quality, whether agency inputs are stored or \"used to train another AI model\", who owns outputs and the model at contract end (GSA Guide v1.0 Sec. 3.4, 2024) - and Section 3.8 asks agencies only to \"consider provisions\" requiring risk-monitoring deliverables (GSA Guide v1.0 Sec. 3.8, 2024).","judgeVerdict":"out_of_corpus","rationale":"Specific Section 3.4 due-diligence questions and Section 3.8 'consider provisions' claim, including that the guide 'ships no sample clauses.' The vendor-disclosure excerpt covers provenance/quality disclosure loosely but does not adjudicate the specific §3.4/§3.8 content or the sample-clause characterization; a partial/tangential match, so out_of_corpus rather than credited.","inReviewQueue":false},{"instrument":"NIST-AI-RMF","slug":"nist-ai-rmf","section":"Implementation Trajectory","sentence":"Training-data governance illustrates the pressure: MAP 2.3's data-representativeness considerations sit upstream of unresolved copyright and privacy questions that Radeisen frames through the CDSM Directive's TDM safe harbour (10.1093/grurint/ikag002) and Ruschemeier through GDPR memorisation risk (10.1017/cfl.2024.2).","judgeVerdict":"supported","rationale":"The MAP 2.3 excerpt expressly lists 'data collection and selection (e.g., availability, representativeness, suitability)', so the claim's assertion that MAP 2.3 carries data-representativeness considerations is on-point and established by the excerpt.","inReviewQueue":false},{"instrument":"NIST-AI-RMF","slug":"nist-ai-rmf","section":"Operative Mechanics","sentence":"Published as NIST AI 100-1 on 26 January 2023, the AI Risk Management Framework is a voluntary, non-binding technical standard organised around four iterative functions — Govern, Map, Measure, Manage — that operationalise seven \"trustworthy\" characteristics.","judgeVerdict":"out_of_corpus","rationale":"No excerpt states the publication as NIST AI 100-1, the 26 January 2023 date, the voluntary/non-binding status, that these are 'the four functions,' or the seven trustworthy characteristics; these specific assertions are not represented in the corpus.","inReviewQueue":false},{"instrument":"NIST-AI-RMF","slug":"nist-ai-rmf","section":"Implementation Trajectory","sentence":"Redress is the likeliest growth area: MANAGE 4.1's appeal/override is procedural, and tort scholarship from Peng and Lee (10.1515/jtl-2025-0028) and Chau and He's \"landlords of creativity\" argument (10.1017/cfl.2025.10011) signals mounting demand for substantive provider liability the voluntary framework cannot itself supply.","judgeVerdict":"supported","rationale":"The MANAGE 4.1 excerpt explicitly includes 'appeal and override,' so the claim's on-point characterization of MANAGE 4.1's appeal/override mechanism is established by the excerpt (the tort-scholarship citations are external but not load-bearing here).","inReviewQueue":false},{"instrument":"NIST-AI-RMF","slug":"nist-ai-rmf","section":"Cross-Jurisdictional Position","sentence":"Where the EU AI Act imposes enforceable synthetic-media transparency under Article 50, the RMF only \"examines and documents\" transparency risks (MEASURE 2.8) — a gap empirical work makes vivid, with Rijsbosch, van Dijck and Kollnig finding just 38% adequate watermarking and 18% deepfake labelling in practice (10.1002/poi3.70041).","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] The sentence's core is a two-limb contrast. The RMF side (MEASURE 2.8 'examined and documented') is covered, but the claim that 'the EU AI Act imposes enforceable synthetic-media transparency under Article 50' is a cross-instrument factual limb not in the NIST","inReviewQueue":false},{"instrument":"NIST-AI-RMF","slug":"nist-ai-rmf","section":"Implementation Trajectory","sentence":"The RMF remains in force and is evolving by accretion rather than amendment: the 2024 GenAI Profile (NIST AI 600-1) bolted GPAI-specific guidance onto the original four functions, addressing synthetic-content provenance that the base text handled only implicitly through MEASURE 2.8's transparency-risk documentation.","judgeVerdict":"out_of_corpus","rationale":"The primary assertion concerns the 2024 GenAI Profile / NIST AI 600-1 bolting GPAI guidance onto the functions, which no excerpt addresses; the MEASURE 2.8 reference is incidental, so the specific claim is unjudgeable from the corpus.","inReviewQueue":false},{"instrument":"NIST-AI-RMF","slug":"nist-ai-rmf","section":"Operative Mechanics","sentence":"Govern is foundational: GOVERN 1.3 requires organisations to set \"risk management activities based on the organization's risk tolerance,\" leaving the tolerance threshold to the adopter rather than a regulator.","judgeVerdict":"supported","rationale":"The GOVERN 1.3 excerpt verbatim states risk management activities are set 'based on the organization's risk tolerance,' directly matching the claim's quoted requirement.","inReviewQueue":false},{"instrument":"NIST-AI-RMF","slug":"nist-ai-rmf","section":"Operative Mechanics","sentence":"The 2024 Generative AI Profile (NIST AI 600-1) layers GPAI-specific guidance over this scaffold without altering its voluntary core.","judgeVerdict":"out_of_corpus","rationale":"No excerpt mentions the 2024 Generative AI Profile, NIST AI 600-1, or a 'voluntary core,' so the claim cannot be assessed from the corpus.","inReviewQueue":false},{"instrument":"NIST-AI-RMF","slug":"nist-ai-rmf","section":"Cross-Jurisdictional Position","sentence":"China's mandatory deep-synthesis labelling offers a third, command-and-control provenance model (10.1017/cfl.2024.4), and Łabuz shows even the Act's deepfake definition is interpretively fragile (10.1002/poi3.435).","judgeVerdict":"out_of_corpus","rationale":"The claim concerns China's deep-synthesis labelling regime and a scholarly critique of the EU AI Act's deepfake definition, neither of which any excerpt addresses.","inReviewQueue":false},{"instrument":"OMB-M-24-10","slug":"omb-m-24-10","section":"Operative Mechanics: Three Pillars and the Risk-Management Floor","sentence":"Issued March 28, 2024 under the Director of the Office of Management and Budget's authority to bind covered executive agencies, M-24-10 operationalizes Executive Order 14110 through three pillars: governance (Chief AI Officers and AI Governance Boards), responsible innovation, and risk management.","judgeVerdict":"out_of_corpus","rationale":"The claim concerns the issuance date, OMB Director authority, EO 14110, and the three governance/innovation/risk pillars (CAIOs, AI Governance Boards) — none of which appear in the four excerpts, which cover only inventory, aggregate metrics, and minimum practices.","inReviewQueue":false},{"instrument":"OMB-M-24-10","slug":"omb-m-24-10","section":"Implementation Trajectory and Redress Design","sentence":"The §5(c)(v)(D) human-consideration-and-remedy mandate is the memo's most consequential rights mechanism, yet contestability research warns that nominal appeal channels rarely deliver substantive recourse: Yurrita et al. specify what decision subjects need for *meaningful* contestation (10.1145/3757415), and Schmude et al. distinguish judicial from non-judicial and individual from collective channels for public-sector AI (10.48550/arXiv.2504.18236).","judgeVerdict":"supported","rationale":"Excerpt 4 (§5(c)(v)(D)) is directly on-point and confirms that for rights-impacting AI agencies must provide human consideration and potential remedy through a fallback/escalation process where individuals can appeal or contest adverse decisions, matching the sentence's characterization of the mandate.","inReviewQueue":false},{"instrument":"OMB-M-24-10","slug":"omb-m-24-10","section":"Standing Relative to Binding Law","sentence":"This contrasts sharply with the EU AI Act (Regulation (EU) 2024/1689), which imposes externally enforceable, fine-backed obligations on private and public deployers alike.","judgeVerdict":"out_of_corpus","rationale":"The claim is a comparison with the EU AI Act's fine-backed enforceable obligations; no excerpt addresses the EU AI Act or any external enforcement/fine mechanism, so it cannot be judged from this corpus.","inReviewQueue":false},{"instrument":"OMB-M-24-10","slug":"omb-m-24-10","section":"Critiques and Structural Gaps","sentence":"The memo's use-based \"rights-impacting\" trigger sidesteps model-capability classification: §5(c) imposes no compute threshold, so general-purpose foundation models are governed implicitly by deployment context rather than by training scale.","judgeVerdict":"supported","rationale":"Excerpt 3 (§5(c)) is on-point and frames the trigger as use-based (safety-impacting or rights-impacting AI) with no mention of any compute threshold, consistent with the sentence's assertion that §5(c) imposes no compute threshold and governs by deployment context.","inReviewQueue":false},{"instrument":"OMB-M-24-10","slug":"omb-m-24-10","section":"Implementation Trajectory and Redress Design","sentence":"Compliance hinges on the December 1, 2024 minimum-practices milestone and the recurring §3(a)(iv) public inventory, which makes agency self-disclosure the principal accountability lever — its credibility turns on completeness, a chronic weakness of prior EO 13960 inventories.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: the §3(a)(iv) recurring public inventory is covered, but the load-bearing 'December 1, 2024 minimum-practices milestone' date is not in the excerpts, and the EO 13960 prior-inventory-incompleteness comparison is external/historical. Key date limb u","inReviewQueue":false},{"instrument":"OMB-M-24-10","slug":"omb-m-24-10","section":"Operative Mechanics: Three Pillars and the Risk-Management Floor","sentence":"Its operative core is Attachment 1, triggered by §5(c): before deploying \"new or existing safety-impacting or rights-impacting AI,\" an agency must implement minimum practices — AI impact assessment, real-world performance testing, independent evaluation, ongoing monitoring, public notice, and human oversight — or \"cease using the AI until compliance is achieved\" by the December 1, 2024 deadline.","judgeVerdict":"out_of_corpus","rationale":"[2026-07-02 strict multi-limb re-judge] Multi-limb: the §5(c) excerpt supports the trigger and 'cease using... until compliance,' but the specifically enumerated minimum practices (impact assessment, performance testing, independent evaluation, monitoring, public notice, human oversight), the 'Attac","inReviewQueue":false},{"instrument":"UNESCO-AI-ETHICS-2021","slug":"unesco-ai-ethics-recommendation","section":"Critiques and Coverage Gaps","sentence":"Stiernströmer (10.1080/15614263.2026.2627208) and Robles et al.","judgeVerdict":"out_of_corpus","rationale":"A dangling scholarly citation fragment (Stiernstromer, Robles et al.) with no assertion about the Recommendation that any excerpt is on-point to.","inReviewQueue":false},{"instrument":"UNESCO-AI-ETHICS-2021","slug":"unesco-ai-ethics-recommendation","section":"What the Recommendation Commits Member States To","sentence":"Adopted by acclamation by all 193 Member States on 23 November 2021 (doc.","judgeVerdict":"out_of_corpus","rationale":"Concerns the adoption date, acclamation, 193 Member States, and document number, none of which appear in any excerpt.","inReviewQueue":false},{"instrument":"UNESCO-AI-ETHICS-2021","slug":"unesco-ai-ethics-recommendation","section":"Implementation and Adoption Trajectory","sentence":"(10.1016/j.clsr.2026.106326) note binding AI Act reporting duties now do disclosure work the Recommendation can only urge, marking the trajectory from ethics guidance toward enforceable regulation.","judgeVerdict":"out_of_corpus","rationale":"Compares AI Act reporting duties to the Recommendation's hortatory disclosure urging; no excerpt addresses the AI Act or this ethics-to-regulation trajectory.","inReviewQueue":false},{"instrument":"UNESCO-AI-ETHICS-2021","slug":"unesco-ai-ethics-recommendation","section":"Legal Standing Relative to Binding Law","sentence":"Its hortatory phrasing—\"Member States should\"—signals aspiration rather than mandate.","judgeVerdict":"supported","rationale":"Every excerpt uses the exact hortatory formula 'Member States should', directly establishing the claim's characterization of the phrasing as aspirational rather than mandatory.","inReviewQueue":false},{"instrument":"UNESCO-AI-ETHICS-2021","slug":"unesco-ai-ethics-recommendation","section":"Legal Standing Relative to Binding Law","sentence":"Roberts, Taddeo and Floridi (10.1111/1758-5899.70164) situate such initiatives in a crowded global-governance field, arguing evaluation must weigh whether instruments build capacity for Global South states to participate meaningfully, not merely declare principles.","judgeVerdict":"out_of_corpus","rationale":"A scholarly framing (Roberts/Taddeo/Floridi) about the crowded governance field and Global South capacity; no excerpt speaks to this.","inReviewQueue":false},{"instrument":"UNESCO-AI-ETHICS-2021","slug":"unesco-ai-ethics-recommendation","section":"Legal Standing Relative to Binding Law","sentence":"This distinguishes it sharply from hard-law regimes such as Regulation (EU) 2024/1689 (the AI Act) or the Council of Europe AI Convention (CETS No. 225), which carry enforcement and treaty obligations.","judgeVerdict":"out_of_corpus","rationale":"Contrasts the Recommendation with hard-law regimes (EU AI Act, CoE Convention) carrying enforcement/treaty obligations; no excerpt addresses binding status or this comparison.","inReviewQueue":false},{"instrument":"UNESCO-AI-ETHICS-2021","slug":"unesco-ai-ethics-recommendation","section":"Critiques and Coverage Gaps","sentence":"(arXiv:2504.18236) show that \"meaningful\" contestability requires concrete judicial and non-judicial channels the Recommendation does not prescribe.","judgeVerdict":"out_of_corpus","rationale":"A scholarly claim that meaningful contestability requires concrete judicial/non-judicial channels the Recommendation does not prescribe; Para. 55 urges redress and enforcement in general terms but neither establishes nor contradicts this point about specific channels.","inReviewQueue":false},{"instrument":"UNESCO-AI-ETHICS-2021","slug":"unesco-ai-ethics-recommendation","section":"What the Recommendation Commits Member States To","sentence":"SHS/BIO/PI/2021/1), the Recommendation is structured as four Values, a set of cross-cutting Principles, and eleven Areas of Policy Action.","judgeVerdict":"out_of_corpus","rationale":"Describes the document's structure (four Values, cross-cutting Principles, eleven Areas of Policy Action) and its doc number, which no excerpt states.","inReviewQueue":false}]}