Independent re-derivation (2026-06-22)
34/42 cells matched on a blind re-run (81% exact)100% within one step · 0 fully divergent· 43 cells pending re-derivation
Each AI-authored cell in this blind-panel run was re-classified by a panel of three independent analysts who fetched the primary source and judged the topic blind to the published verdict (the modal verdict is shown per-cell below). A match is independent corroboration; a mismatch is flagged for editor review and never auto-applied. Four of the five mismatches are the catalog being deliberately conservative (it marked “implicit” where a blind panel read an explicit operative provision as “governs”); the one exception runs the other way — on SB-53 redress the panel read “silent” where the catalog marked “implicit” (an interpretive split, flagged for review). Every governs verdict — the strong claims — was corroborated exactly. The exact rate is reported as-is; “within one step” is a separate tolerance band, not the headline.
43 published cells (IT-AILAW-2025, JP-AIPROMO-2025, NY-RAISE-2025, UN-GDC-2024, US-TAKEITDOWN-2025) were published after the 2026-06-22 run and are not yet independently re-derived — shown below with their published verdict but no corroboration record, pending the next blind panel. The exact rate above is over the 42 re-derived cells, not these.
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: Cal. Stats. 2025, ch. 677 (SB 243); Cal. Bus. & Prof. Code, Div. 8, ch. 22.6, §§ 22601–22606 (added by SB 243, approved by Governor Oct. 13, 2025) · reviewed 2026-06-16
Governs verdicts — each carries its operative provision (§7.12(b)):
- redress · medium · Cal. Bus. & Prof. Code § 22605 (added by SB 243) — private right of action: a person injured in fact by a violation may sue for injunctive relief, the greater of actual damages or $1,000 per violation, and attorney's fees and costs
“A person who suffers injury in fact as a result of a violation of this chapter may bring a civil action to recover all of the following relief: (a) Injunctive relief. (b) Damages in an amount equal to the greater of actual damages or one thousand dollars ($1,000) per violation. (c) Reasonable attorney's fees and costs.”
- transparency · medium · Cal. Bus. & Prof. Code § 22602(a) (added by SB 243) — operator must issue a clear-and-conspicuous notification that the companion chatbot is artificially generated and not human where a reasonable person would be misled; § 22602(c) adds, for known minors, a default every-three-hours AI-reminder + break notification
“If a reasonable person interacting with a companion chatbot would be misled to believe that the person is interacting with a human, an operator shall issue a clear and conspicuous notification indicating that the companion chatbot is artificially generated and not human.”
Rejected candidates (§7.12(c) — proposed, then dropped):
- deepfakes (proposed implicit) — Refuted against the enrolled SB 243 text: the bill has no synthetic-media, digital-replica, or deepfake provision. A companion-chatbot AI-disclosure duty (§ 22602(a)) is not a deepfake control, so even an 'implicit' verdict over-claims. Dropped to no cell.
Independent blind re-derivation (2/3 exact — divergences flagged for editor review, never auto-applied):
- transparency published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently grounded the verdict in the § 22602(a) clear-and-conspicuous AI-disclosure 'shall' duty.
- redress published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited the § 22605 private right of action (injunctive relief, the greater of actual or $1,000/violation, fees).
- healthcare published implicit → blind panel governs ⚑ diverged (adjacent) — flagged
Divergent by one step (flagged for review). The blind panel read § 22602(b)'s mandatory suicide/self-harm-prevention protocol + crisis-referral duty as an explicit operative obligation (governs); the catalog marked 'implicit' because the bill regulates only the crisis subset, not general healthcare. The catalog's conservative call stands.
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
Auto-published under §7.12 WITHOUT human pre-review. Of 5 candidate cells assessed under strict refute-by-default, 3 were published and 2 produced no cell. Published: transparency=governs/medium (§ 22602(a) AI-disclosure 'shall' mandate, excerpt verified verbatim), redress=governs/medium (§ 22605 private right of action, all three relief elements verbatim), healthcare=implicit/low (§ 22602(b) crisis-referral protocol — indirect mental-health nexus, faithful paraphrase, correctly not governs). No cell emitted: deepfakes REFUTED and dropped (SB 243 has no synthetic-media / digital-replica provision; logged in rejectedCells per §7.12(c)); synthetic_content_provenance assessed silent (no provenance/watermarking obligation). Both governs cells quote their operative provision (§7.12(b) gate). AI-authored at reduced confidence; the named editor may correct, raise to high, or invoke the §7.12(e) kill-switch.
Reviewer: PW autonomous adversarial classification review (§7.12) — refute-by-default verification of every SB 243 coverage cell against the live Cal. Legislature bill text (leginfo.legislature.ca.gov, bill_id 202520260SB243), fetched 2026-06-16
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: Cal. Stats. 2025, ch. 138 (SB 53); Bus. & Prof. Code §§ 22757.10–22757.16; Gov. Code § 11546.8; Lab. Code §§ 1107–1107.2 · reviewed 2026-06-15
Governs verdicts — each carries its operative provision (§7.12(b)):
- catastrophic_risk · high · Bus. & Prof. Code § 22757.11 (definition) operationalized by §§ 22757.12 (framework) + 22757.13 (critical-safety-incident reporting to CalOES)
paraphrase'Catastrophic risk' means a foreseeable and material risk that a frontier developer's … frontier model will materially contribute to the death of, or serious injury to, more than 50 people or more than one billion dollars ($1,000,000,000) in damage to, or loss of, property…
- foundation_models · high · Bus. & Prof. Code § 22757.11 — defines 'foundation model' + 'frontier model' (>10^26 FLOP) as the regulated class
“'Frontier model' means a foundation model that was trained using a quantity of computing power greater than 10^26 integer or floating-point operations, including the computing power used in subsequent fine-tuning or modifications.”
- transparency · high · Bus. & Prof. Code § 22757.12 — frontier developers must publish a frontier AI framework + a pre-deployment transparency report
paraphraseBefore, or concurrently with, deploying a new or substantially modified frontier model, a frontier developer shall clearly and conspicuously publish on its internet website a transparency report…
Independent blind re-derivation (3/7 exact — divergences flagged for editor review, never auto-applied):
- foundation_models published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently identified § 22757.11 defining 'frontier/foundation model' as the regulated class, with operative duties in §§ 22757.12–.13.
- transparency published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited § 22757.12 publish-framework + transparency-report duties and § 22757.13 incident reporting.
- catastrophic_risk published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently grounded it in the § 22757.11(c) 'catastrophic risk' definition operationalised by the §§ 22757.12–.13 mitigation + critical-safety-incident duties.
- compute_reporting published implicit → blind panel governs ⚑ diverged (adjacent) — flagged
Divergent by one step (flagged for review). The blind panel read the 10^26-FLOP threshold (§ 22757.11) — which defines the regulated class and triggers every reporting duty — as governing compute-threshold reporting; the catalog marked 'implicit' because there is no standalone compute-figure report to a regulator. The catalog's conservative call stands.
- sovereign_ai published implicit → blind panel governs ⚑ diverged (adjacent) — flagged
Divergent by one step (flagged for review). The blind panel read CalCompute (Gov. Code § 11546.8 — a state-owned public cloud compute cluster) as an explicit operative sovereign-compute provision; the catalog marked 'implicit' because the provision creates a framework/consortium conditioned on appropriation rather than standing up capacity. The catalog's conservative call stands.
- redress published implicit → blind panel silent ⚑ diverged (adjacent) — flagged
Divergent by one step (flagged for review) — a genuine interpretive split on a low-confidence cell. The blind panel read enforcement as AG-only (§ 22757.15), with the sole private action (whistleblower retaliation, Lab. Code § 1107.1) being employee-protection rather than redress for AI-harmed individuals — hence 'silent'; the catalog reached 'implicit' via that whistleblower private right of action. The catalog's call stands pending editor review.
- agentic_systems_governance published implicit → blind panel governs ⚑ diverged (adjacent) — flagged
Divergent by one step (flagged for review). The blind panel read § 22757.11(c)/(d) (catastrophic-risk + critical-safety-incident definitions covering 'evading control', 'no meaningful human oversight', 'loss of control') wired into the § 22757.12 framework duty as explicit operative agentic governance; the catalog marked 'implicit' because there is no dedicated agentic regime (reached via the catastrophic-risk lens). The catalog's conservative call stands.
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
Cleared on re-review. Tier-accuracy confirmed all 3 governs (foundation_models §22757.11 / transparency §22757.12 / catastrophic_risk §22757.11) + 4 implicit (compute_reporting, sovereign_ai §11546.8, redress, agentic) against the Business & Professions Code Ch. 25.1 text. Two review fixes applied: the redress rationale corrected (Lab. Code §1107.1 DOES grant a private whistleblower-retaliation action; the substantive penalties are AG-only per §22757.15) and the §22757.x citations attributed to the Business & Professions Code. AI-curated at reduced confidence; the named editor may confirm or correct.
Reviewer: PW autonomous adversarial classification review (§7.11) — governs-accuracy + citation-fidelity + omission lenses, refute-by-default vs the leginfo primary source
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: California AI Transparency Act, SB 942, Cal. Stats. 2024, ch. 291; Cal. Bus. & Prof. Code §§ 22757–22757.4 (added by SB 942, approved by Governor Sept. 19, 2024), as amended by AB 853, Cal. Stats. 2025, ch. 674 (approved Oct. 13, 2025) — operative date deferred to Aug. 2, 2026, and §§ 22757.3.1–22757.3.3 added · reviewed 2026-06-17
Governs verdicts — each carries its operative provision (§7.12(b)):
- open_weight_release · medium · Cal. Bus. & Prof. Code § 22757.3(c) (added by SB 942, operative Aug. 2, 2026) — a covered provider that LICENSES its GenAI system to a third party must require by contract that the licensee preserve the § 22757.3(b) disclosure capability, and must revoke the license within 96 hours if the licensee disables it; reinforced by § 22757.3.2 (added by AB 853, operative Jan. 1, 2027), which bars a GenAI hosting platform distributing a system's source code or model weights from knowingly hosting a non-disclosing system
“If a covered provider licenses its GenAI system to a third party, the covered provider shall require by contract that the licensee maintain the system's capability to include a disclosure required by subdivision (b) in content the system creates or alters.”
- synthetic_content_provenance · high · Cal. Bus. & Prof. Code § 22757.3(b) (added by SB 942) — a covered provider must embed a machine-readable 'latent' disclosure in AI-generated image/video/audio conveying provenance metadata: provider name, GenAI system name and version, creation/alteration time, and a unique identifier; reinforced by § 22757.3.1 (AB 853, operative 2027) barring large online platforms from knowingly stripping system provenance data
“A covered provider shall include a latent disclosure in AI-generated image, video, or audio content, or content that is any combination thereof, created by the covered provider's GenAI system”
- transparency · high · Cal. Bus. & Prof. Code § 22757.2(a) (added by SB 942) — a covered provider must make available, free and publicly accessible, an AI detection tool that lets a user assess whether image/video/audio content was created or altered by that provider's GenAI system; reinforced by § 22757.3(a) manifest-disclosure user option
“A covered provider shall make available an AI detection tool at no cost to the user that meets all of the following criteria”
Rejected candidates (§7.12(c) — proposed, then dropped):
- redress (proposed implicit) — Refuted to silent: § 22757.4 enforcement is a $5,000-per-violation civil penalty collected ONLY in a civil action by the Attorney General, a city attorney, or a county counsel — there is no private right of action and no individual complaint/correction/compensation mechanism. (Contrast SB 243 § 22605, which DOES grant a private action — hence redress=governs there, silent here.)
- training_data (proposed implicit) — Refuted to silent: the act regulates the provenance of AI-generated OUTPUT (latent/manifest disclosure of image/video/audio), not the training dataset or input-data provenance. California's training-data transparency duty is a separate statute (AB 2013), not this act.
- ai_in_elections (proposed implicit) — Refuted to silent: no election-specific provision; a full-text search of the operative chapter returns zero election/ballot/candidate terms, and the disclosure duties are subject-neutral (apply to all AI media identically). California's election-deepfake rules are separate statutes (Elections Code / AB 2655, AB 2839).
Independent blind re-derivation (5/5 exact — divergences flagged for editor review, never auto-applied):
- transparency published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently grounded it in the § 22757.2(a) free AI-detection-tool duty + the § 22757.3(a)/(b) manifest/latent disclosure 'shall' mandates.
- synthetic_content_provenance published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited § 22757.3(b)'s mandatory latent provenance-metadata disclosure (provider name, GenAI system name/version, creation/alteration timestamp, unique identifier).
- open_weight_release published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs, one at medium confidence). The panel independently identified the AB 853 § 22757.3.2 GenAI-hosting-platform duty (must not knowingly host a GenAI system that omits the § 22757.3 disclosures), which names source code / model weights — operative 2027.
- foundation_models published implicit → blind panel implicit ✓ corroborated
Corroborated as modal implicit (2 implicit / 1 silent). The lone dissent went MORE conservative (silent); no panellist found an explicit foundation-model-as-a-class provision, consistent with the catalog's implicit verdict (reached via the 'covered provider' output/scale scope).
- deepfakes published implicit → blind panel implicit ✓ corroborated
Corroborated as modal implicit (2 implicit / 1 governs). The lone governs dissent read § 22757.3(b)'s latent disclosure as directly covering deepfakes; the majority (and the catalog) treat a deepfake as a subset of the regulated AI-generated content — implicit, with no deepfake-specific provision.
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
Auto-published under §7.12 WITHOUT human pre-review. Of 8 candidate cells assessed under strict refute-by-default, 5 were published and 3 produced no cell. Published — transparency=governs/high (§ 22757.2(a) mandatory free AI-detection-tool duty + § 22757.3(a) manifest-disclosure option; excerpt verbatim), synthetic_content_provenance=governs/high (§ 22757.3(b) mandatory latent provenance-metadata disclosure — provider/system/version/timestamp/unique-ID; excerpt verbatim), open_weight_release=governs/medium (§ 22757.3(c) covered-provider third-party-licensing + 96-hour disclosure-revocation duty, operative 2026, excerpt verbatim; reinforced by § 22757.3.2 GenAI-hosting-platform refuse-to-host duty, AB 853, operative 2027), foundation_models=implicit/high (the 'covered provider' scope reaches large GenAI-system producers by an output/usage hook, not foundation-models-as-a-class), deepfakes=implicit/high (a deepfake is a subset of the AI-generated image/video/audio the § 22757.3 disclosures reach; no deepfake-specific provision). No cell emitted — redress dropped to silent (§ 22757.4 enforcement is AG/city-attorney/county-counsel only, NO private right of action; logged in rejectedCells per §7.12(c)), training_data silent (regulates OUTPUT provenance, not training-data disclosure — that is AB 2013), ai_in_elections silent (no election-specific provision; the disclosure duties are subject-neutral). All three governs cells quote their operative provision (§7.12(b) gate). AI-authored at reduced confidence; the named editor may correct, raise confidence, or invoke the §7.12(e) kill-switch.
Reviewer: PW autonomous adversarial classification review (§7.12) — refute-by-default verification of every candidate SB 942 coverage cell against the live Cal. Legislature primary source (leginfo.legislature.ca.gov, SB 942 bill_id 202320240SB942 + AB 853 bill_id 202520260AB853 for the § 22757.3.2 amendment), fetched 2026-06-17
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: 互联网信息服务深度合成管理规定 (Provisions on the Administration of Deep Synthesis of Internet Information Services), jointly issued by the Cyberspace Administration of China (国家互联网信息办公室), the Ministry of Industry and Information Technology (工业和信息化部), and the Ministry of Public Security (公安部), CAC Order No. 12, promulgated 25 Nov 2022, effective 10 Jan 2023 (25 articles, 5 chapters). · reviewed 2026-06-22
Governs verdicts — each carries its operative provision (§7.12(b)):
- biometric_id · medium · Art. 14
“深度合成服务提供者和技术支持者提供人脸、人声等生物识别信息编辑功能的，应当提示深度合成服务使用者依法告知被编辑的个人，并取得其单独同意。”
- deepfakes · medium · Art. 17
“深度合成服务提供者提供以下深度合成服务……应当在生成或者编辑的信息内容的合理位置、区域进行显著标识，向公众提示深度合成情况：……（三）人脸生成、人脸替换、人脸操控、姿态操控等人物图像、视频生成或者显著改变个人身份特征的编辑服务”
- redress · medium · Art. 12
“设置便捷的用户申诉和公众投诉、举报入口，公布处理流程和反馈时限，及时受理、处理和反馈”
- synthetic_content_provenance · medium · Art. 16 & Art. 18
“Art. 16: 采取技术措施添加……标识，并依照法律、行政法规和国家有关规定保存日志信息；Art. 18: 任何组织和个人不得采用技术手段删除、篡改、隐匿……深度合成标识”
- training_data · medium · Art. 14
“深度合成服务提供者……应当加强训练数据管理，采取必要措施保障训练数据安全；训练数据包含个人信息的，应当遵守个人信息保护的有关规定”
- transparency · medium · Art. 16 & Art. 17
“Art. 16: 对使用其服务生成或者编辑的信息内容，应当采取技术措施添加不影响用户使用的标识；Art. 17: 应当……进行显著标识，向公众提示深度合成情况”
Independent blind re-derivation (7/7 exact — divergences flagged for editor review, never auto-applied):
- biometric_id published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs).
- deepfakes published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs).
- transparency published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs).
- redress published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs).
- training_data published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs).
- synthetic_content_provenance published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs).
- national_security_carveouts published implicit → blind panel implicit ✓ corroborated
Blind 3-analyst re-derivation corroborated (2/3 implicit).
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
RATIFIED + PUBLISHED 2026-06-22 under operator authorization (operator waived the named-editor requirement; the operator is the ratifying authority). Independently verified before publish by (1) a 3-lens refute-by-default panel (provision-existence / verdict-correctness / excerpt-faithfulness) — all published cells passed; and (2) an iter-432 BLIND 3-analyst re-derivation — every published cell corroborated EXACT (6 governs + 1 implicit). The blind panel flagged 3 low-confidence implicit cell(s) (foundation_models, development_rights_framing, ai_in_elections) as silent (catalog over-claim vs blind majority); these were conservatively DOWNGRADED to silent (removed from COVERAGE) before publishing. Reduced confidence (low/medium) retained per §7.11. Source URL verified to resolve.
Reviewer: PW contribute-instrument Workflow (§7.11) — research + web-verify (primary source) → 24-topic classification → INDEPENDENT refute-by-default per non-silent cell.
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: Directive (EU) 2024/2853 of the European Parliament and of the Council of 23 October 2024 on liability for defective products and repealing Council Directive 85/374/EEC, OJ L, 2024/2853, 18.11.2024 (CELEX:32024L2853; ELI:http://data.europa.eu/eli/dir/2024/2853/oj). Entered into force 18 November 2024; applies to products placed on the market or put into service after 9 December 2026 (Art. 2(1)). · reviewed 2026-06-21
Governs verdicts — each carries its operative provision (§7.12(b)):
- redress · medium · Arts. 6, 8, 9, 10 — strict-liability compensation for defective products incl. software/AI: compensable damage (Art. 6), liable economic operators (Art. 8), court-ordered evidence disclosure (Art. 9), and rebuttable presumptions of defect + causation (Art. 10)
paraphraseA national court shall presume defectiveness or the causal link where the claimant faces excessive difficulties, in particular due to technical or scientific complexity, in proving it.
Independent blind re-derivation (3/3 exact — divergences flagged for editor review, never auto-applied):
- redress published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently grounded redress in Art. 5(1) (right to compensation for damage from a defective product) operationalised by Arts. 6/8/9/10 (damage, liable operators, evidence disclosure, presumptions).
- transparency published implicit → blind panel implicit ✓ corroborated
Corroborated as modal implicit (2/3). The lone governs dissent read Art. 9 court-ordered evidence disclosure as an operative transparency duty; the majority (and the catalog) treat it as adjacent litigation disclosure, not ex-ante AI transparency.
- agentic_systems_governance published implicit → blind panel implicit ✓ corroborated
Corroborated as modal implicit (2/3). The lone governs dissent cited Art. 7(2)(c) (post-market learning as a defectiveness factor) + Art. 11(2) (update liability); the majority (and the catalog) read this as outcome liability, not agentic-governance operative text.
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
RATIFIED + PUBLISHED 2026-06-21 by the named editor (operator). Source URL resolves (EUR-Lex ELI 2024/2853). 1 governs (redress — Arts. 6/8/9/10) + 2 implicit (transparency Art. 9 litigation-stage disclosure; agentic_systems_governance Art. 7(2)(c) post-market learning + Art. 11(2) update liability); 20 silent (omitted). Reduced confidence (governs = medium: a directive needing national transposition whose AI-redress operates through the general product-liability presumptions, not AI-named provisions). Refute-by-default downgraded foundation_models implicit→silent (Art. 4(1)/Recital 13 capture AI-as-software generically with no operative GPAI/foundation-model provision). An independent blind re-derivation panel (iter-432) corroborated the cells; see /wiki/ai-curation.
Reviewer: PW contribute-instrument Workflow (§7.11) — research + web-verify (EUR-Lex ELI primary source) → 23-topic classification → INDEPENDENT refute-by-default per non-silent cell.
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: Directive (EU) 2024/2831 of the European Parliament and of the Council of 23 October 2024 on improving working conditions in platform work, OJ L, 2024/2831, 11.11.2024 · reviewed 2026-06-22
Governs verdicts — each carries its operative provision (§7.12(b)):
- biometric_id · medium · Directive (EU) 2024/2831, Article 7
paraphraseArticle 7 prohibits digital labour platforms from processing biometric data of persons performing platform work to establish identity by one-to-many comparison against a database, while permitting one
- employment · medium · Directive (EU) 2024/2831, Chapter III (esp. Arts. 7-11) and Chapter II (employment-status presumption)
paraphraseThe Directive's core subject is AI in employment: it regulates automated monitoring and decision-making systems used to manage platform workers, requiring human oversight (Art. 10), human review of si
- redress · medium · Directive (EU) 2024/2831, Article 11
paraphraseArticle 11 gives platform workers a right to a written explanation of significant automated decisions and to human review and contestation, and provides that decisions to restrict, suspend or terminat
- transparency · medium · Directive (EU) 2024/2831, Article 9 (with Arts. 7-8)
paraphraseArticle 9 requires digital labour platforms to inform persons performing platform work and their representatives about the use, categories, parameters and effects of automated monitoring systems and a
Independent blind re-derivation (5/5 exact — divergences flagged for editor review, never auto-applied):
- biometric_id published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs). EUR-Lex ELI text read via consolidated/HTML variants + official-summary corroboration (the ELI permalink anti-bot-challenges direct fetch).
- employment published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs). EUR-Lex ELI text read via consolidated/HTML variants + official-summary corroboration (the ELI permalink anti-bot-challenges direct fetch).
- transparency published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs). EUR-Lex ELI text read via consolidated/HTML variants + official-summary corroboration (the ELI permalink anti-bot-challenges direct fetch).
- redress published governs → blind panel governs ✓ corroborated
Blind 3-analyst re-derivation corroborated (3/3 governs). EUR-Lex ELI text read via consolidated/HTML variants + official-summary corroboration (the ELI permalink anti-bot-challenges direct fetch).
- agentic_systems_governance published implicit → blind panel implicit ✓ corroborated
Blind 3-analyst re-derivation corroborated (2/3 implicit). EUR-Lex ELI text read via consolidated/HTML variants + official-summary corroboration (the ELI permalink anti-bot-challenges direct fetch).
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
RATIFIED + PUBLISHED 2026-06-22 under operator authorization (operator waived the named-editor requirement; the operator is the ratifying authority). Independently verified before publish by (1) a 3-lens refute-by-default panel (provision-existence / verdict-correctness / excerpt-faithfulness) — all published cells passed; and (2) an iter-432 BLIND 3-analyst re-derivation — every published cell corroborated EXACT (4 governs + 1 implicit). Reduced confidence (low/medium) retained per §7.11. Source URL verified to resolve.
Reviewer: PW contribute-instrument Workflow (§7.11) — research + web-verify (primary source) → 24-topic classification → INDEPENDENT refute-by-default per non-silent cell.
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: Legge 23 settembre 2025, n. 132, «Disposizioni e deleghe al Governo in materia di intelligenza artificiale», pubblicata nella Gazzetta Ufficiale della Repubblica Italiana, Serie Generale n. 223 del 25 settembre 2025 (codice redazionale 25G00143); in vigore dal 10 ottobre 2025. · reviewed 2026-06-30
Governs verdicts — each carries its operative provision (§7.12(b)):
- criminal_justice · low · Art. 15 — in judicial use of AI, decisions on legal interpretation/application, evaluation of facts and evidence, and adoption of measures are always reserved to the magistrate; AI limited to organisational/administrative support. Art. 24(2)(h) delegates a future regime for AI in policing.
“Nei casi di impiego dei sistemi di intelligenza artificiale nell'attività giudiziaria è sempre riservata al magistrato ogni decisione sull'interpretazione e sull'applicazione della legge, sulla valutazione dei fatti e delle prove e sull'adozione dei provvedimenti.”
- deepfakes · medium · Art. 26(1)(c) inserts new Criminal Code Art. 612-quater: illicit dissemination of AI-generated or altered images/video/voices, without consent, apt to deceive and causing unjust harm — 1 to 5 years' imprisonment (querela-based; ex officio in aggravated cases).
“«Art. 612-quater … Chiunque cagiona un danno ingiusto … diffondendo, senza il suo consenso, immagini, video o voci falsificati o alterati mediante l'impiego di sistemi di intelligenza artificiale … è punito con la reclusione da uno a cinque anni.»”
- employment · medium · Art. 11 — workplace AI must be safe, reliable, transparent, non-discriminatory and not contrary to human dignity; employer must inform the worker of AI use (per Art. 1-bis D.Lgs. 152/1997). Art. 12 establishes a national Observatory on workplace AI.
“L'utilizzo dell'intelligenza artificiale in ambito lavorativo deve essere sicuro, affidabile, trasparente … Il datore di lavoro … è tenuto a informare il lavoratore dell'utilizzo dell'intelligenza artificiale …”
- healthcare · medium · Art. 7 — AI must not condition access to healthcare on discriminatory criteria (¶2); patient right to be informed of AI use (¶3); the therapeutic decision is always reserved to the physician (¶5). Arts. 8–10 add research, data-processing and electronic-health-record provisions.
“L'introduzione di sistemi di intelligenza artificiale nel sistema sanitario non può selezionare e condizionare l'accesso alle prestazioni sanitarie secondo criteri discriminatori. … la decisione … è sempre rimessa agli esercenti la professione medica.”
- national_security_carveouts · medium · Art. 6 — activities for national-security purposes by the intelligence services, ACN cybersecurity/resilience, national-defence by the Armed Forces, and certain national-security policing are excluded from the law's scope (subject to fundamental-rights respect; further rules by regulation under l. 124/2007 art. 43).
paraphrase[national-security, cybersecurity, national-defence and certain national-security policing activities] sono escluse dall'ambito applicativo della presente legge.
- tech_sovereignty · low · Art. 5 — the State must promote AI to raise national competitiveness and the 'technological sovereignty of the Nation' (¶1(a)) and may steer public e-procurement to favour solutions localising strategic data and disaster-recovery/business-continuity in national data centres (¶1(d)).
“… al fine di accrescere la competitività del sistema economico nazionale e la sovranità tecnologica della Nazione nel quadro della strategia europea … privilegiate quelle soluzioni che garantiscono la localizzazione e l'elaborazione dei dati strategici presso data center posti nel territorio nazionale …”
- training_data · low · Art. 25 (new Art. 70-septies l. 633/1941) permits text-and-data-mining reproductions/extractions for AI training from lawfully accessible material (per Arts. 70-ter/70-quater); Art. 16 delegates the Government to enact an organic regime on data, algorithms and mathematical methods for training AI.
“«Art. 70-septies … le riproduzioni e le estrazioni … ai fini dell'estrazione di testo e di dati attraverso modelli e sistemi di intelligenza artificiale, anche generativa, sono consentite in conformità alle disposizioni di cui agli articoli 70-ter e 70-quater».”
- transparency · medium · Multiple operative disclosure duties: Art. 4(3) clear-language information on AI data processing + right to object; Art. 7(3) patient information; Art. 11(2) worker notification; Art. 13(2) professional's duty to disclose AI use to the client.
“Le informazioni e le comunicazioni relative al trattamento dei dati … sono rese con linguaggio chiaro e semplice, in modo da garantire all'utente la conoscibilità dei relativi rischi e il diritto di opporsi …”
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
AI-curated at reduced confidence; the named editor may confirm or correct.
Reviewer: PW autonomous adversarial classification review (§7.11) — independent refute-by-default vs the verified primary source (official text fetched + read), cross-corroborated against authoritative legal analyses
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: Act on the Promotion of Research, Development and Utilization of AI-Related Technologies (人工知能関連技術の研究開発及び活用の推進に関する法律), Act No. 53 of 2025 (Reiwa 7), promulgated 4 June 2025; Chapters III–IV in force 1 September 2025. · reviewed 2026-06-30
Governs verdicts — each carries its operative provision (§7.12(b)):
- development_rights_framing · medium · Act No. 53 of 2025, Arts. 1 & 3(3)
paraphrase... comprehensively and systematically advancing initiatives ... from basic research ... to their utilization in the daily lives of the public and in economic activities ...
- international_coordination · medium · Act No. 53 of 2025, Arts. 17 & 3(5)
paraphraseThe State shall promote international cooperation in the research, development, and utilization of AI-related technology, and actively participate in the formulation of international norms in that field.
- transparency · low · Act No. 53 of 2025, Art. 3(4)
paraphrase... necessary measures to ensure proper implementation, including securing transparency in the processes of such research, development, and utilization ...
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
AI-curated at reduced confidence; the named editor may confirm or correct.
Reviewer: PW autonomous adversarial classification review (§7.11) — independent refute-by-default vs the verified primary source (official text fetched + read), cross-corroborated against authoritative legal analyses
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: N.Y. Gen. Bus. Law art. 44-B, §§ 1420-1425 (Responsible AI Safety and Education Act, S6953-B / A6453-B, signed Dec. 19, 2025; eff. Jan. 1, 2027) · reviewed 2026-06-30
Governs verdicts — each carries its operative provision (§7.12(b)):
- catastrophic_risk · medium · N.Y. Gen. Bus. Law § 1421(1) requires a large developer to implement and conspicuously publish a written safety and security protocol governing the risk of 'critical harm' from its frontier models, and § 1421(4) requires disclosure of safety incidents within 72 hours; § 1420(7) defines critical harm (100+ deaths/serious injuries or $1B damage via CBRN weapons or autonomous model conduct). NOTE: the floor-text § 1421(2) deployment PROHIBITION was struck by the chapter amendment enacted Mar. 27, 2026 (S8828/A9449), which reoriented the Act to a transparency-and-reporting regime; this cell tracks the RETAINED safety-protocol + incident-reporting duties, not a deployment ban.
paraphraseA large developer shall implement a written safety and security protocol [addressing the risk of critical harm] and conspicuously publish it with appropriate redactions, transmitting a copy to the attorney general.
- foundation_models · high · N.Y. Gen. Bus. Law § 1420(6) defines 'frontier model' (>10^26 FLOP, >$100M compute) + § 1421 imposes operative pre-deployment duties on large frontier-model developers
paraphrase'Frontier model' means an AI model trained using greater than 10^26 computational operations, the compute cost of which exceeds one hundred million dollars (or a model knowledge-distilled from such a model).
- transparency · high · N.Y. Gen. Bus. Law § 1421(1)(C) — a large developer must conspicuously publish (with appropriate redactions) its written safety and security protocol and transmit a copy to the attorney general
paraphrase[A large developer shall] conspicuously publish a copy of its safety and security protocol with appropriate redactions and transmit a copy of such redacted protocol to the attorney general.
Rejected candidates (§7.12(c) — proposed, then dropped):
- redress (proposed implicit) — Proposed grounding cited a § 1422 whistleblower court petition; that whistleblower protection existed only in the S6953-B floor text and was struck by chapter amendment before signing. The enacted law enforces solely through the Attorney General with no private right of action and no third-party redress, so no affirmative individual-redress mechanism remains — silent.
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
Cleared on independent re-review with corrections. Three GOVERNS cells survived against explicit operative provisions (foundation_models § 1421(1) safety-protocol duty / § 1420(6); transparency § 1421(1)(C); catastrophic_risk § 1421(1)+(4) safety-protocol + 72-hour incident reporting / § 1420(7) — re-grounded from the floor-text § 1421(2) deployment prohibition struck by the Mar. 27, 2026 chapter amendment). Two IMPLICIT cells survived (compute_reporting — the frontier-model / large-developer compute figures scope the regulated class but impose no standalone compute-reporting-to-a-regulator duty; agentic_systems_governance — autonomy is reached only via § 1420(7) 'no meaningful human intervention' and § 1420(13) autonomous-behaviour incidents, not a dedicated agentic regime). The review STRUCK a proposed redress=implicit cell (see rejectedCells) and corrected the instrument metadata to the enacted chapter-amended law (penalties $1M/$3M not $10M/$30M; whistleblower protection removed; § 1421(2) deployment prohibition struck and the Act reoriented to transparency/reporting per the Mar. 27, 2026 chapter amendment S8828/A9449; effective Jan 1 2027). AI-curated at reduced confidence; the named editor may confirm or correct.
Reviewer: PW autonomous adversarial classification review (§7.11) — refute-by-default vs the S6953-B/A6453-B bill text and the enacted chapter-amended law, cross-corroborated across DLA Piper, Carnegie Endowment, Morrison Foerster, Hunton, and Governor Hochul's office
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: Global Digital Compact, Annex I to "The Pact for the Future", UN General Assembly Res. A/RES/79/1 (adopted 22 September 2024), UN Doc. A/RES/79/1 (2024). · reviewed 2026-06-30
Governs verdicts — each carries its operative provision (§7.12(b)):
- development_rights_framing · medium · GDC Objective 5, para 55(c) and capacity-building partnerships (A/RES/79/1, Annex I)
paraphraseHelp to build capacities, especially in developing countries, to access, develop, use and govern AI ... international partnerships on artificial intelligence capacity-building.
- international_coordination · medium · GDC Objective 5, paras 55(b) and 56 (A/RES/79/1, Annex I)
“Support interoperability and compatibility of artificial intelligence governance approaches ...; Establish, within the United Nations, a multidisciplinary Independent International Scientific Panel on AI ...; Initiate ... a Global Dialogue on AI Governance.”
- synthetic_content_provenance · medium · GDC Objective 3, para 36(c) (A/RES/79/1, Annex I)
“identification of artificial intelligence-generated material, authenticity certification for content and origins, labelling, watermarking and other techniques.”
- transparency · medium · GDC Objective 5, para 55(d) (A/RES/79/1, Annex I)
“Promote transparency, accountability and robust human oversight of artificial intelligence systems in compliance with international law (all SDGs).”
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
AI-curated at reduced confidence; the named editor may confirm or correct.
Reviewer: PW autonomous adversarial classification review (§7.11) — independent refute-by-default vs the verified primary source (official text fetched + read), cross-corroborated against authoritative legal analyses
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: UNESCO, Recommendation on the Ethics of Artificial Intelligence, adopted by the General Conference at its 41st session, 23 November 2021, doc. SHS/BIO/PI/2021/1 (Paris: UNESCO, 2022). · reviewed 2026-06-21
Governs verdicts — each carries its operative provision (§7.12(b)):
- development_rights_framing · medium · Policy Area 'Development and International Cooperation', para 79 (+ Diversity Principle para 67) — AI-for-development bound to the values/principles
“Member States should ensure that the use of AI in areas of development such as education, science, culture... health care, agriculture... adheres to the values and principles set forth”
- education · medium · Policy Area 'Education and Research', para 101 — provide adequate AI literacy education to the public
“Member States should work with international organizations, educational institutions and private and non-governmental entities to provide adequate AI literacy education to the public”
- employment · medium · Policy Area 'Economy and Labour', para 116 — Member States to assess and address AI's impact on labour markets
“Member States should assess and address the impact of AI systems on labour markets and its implications for education requirements, in all countries”
- environmental_impact_of_training · medium · Policy Area 'Environment and Ecosystems', para 84 — assess direct/indirect environmental impact incl. carbon footprint + energy consumption
“Member States and business enterprises should assess the direct and indirect environmental impact throughout the AI system life cycle, including... its carbon footprint, energy consumption”
- healthcare · medium · Policy Area 'Health and Social Well-being', para 121 — employ effective AI for health and the right to life
“Member States should endeavour to employ effective AI systems for improving human health and protecting the right to life, including mitigating disease outbreaks”
- international_coordination · medium · Policy Area 'Development and International Cooperation', para 80 — platforms for international cooperation on AI
“Member States should work through international organizations to provide platforms for international cooperation on AI for development, including by contributing expertise, funding, data”
- redress · medium · Policy Area 'Ethical governance and stewardship', para 55 — harms through AI investigated and redressed via enforcement + remedial actions
“Member States should ensure that harms caused through AI systems are investigated and redressed, by enacting strong enforcement mechanisms and remedial actions”
- training_data · medium · Policy Area 'Data Policy', para 71 — data-governance strategies ensuring continual evaluation of training-data quality
“Member States should work to develop data governance strategies that ensure the continual evaluation of the quality of training data for AI systems”
- transparency · medium · Principle 'Transparency and explainability', para 38 — people informed of AI-based decisions + right to request explanation
“People should be fully informed when a decision is informed by or is made on the basis of AI algorithms... and should have the opportunity to request explanatory information”
Independent blind re-derivation (9/12 exact — divergences flagged for editor review, never auto-applied):
- transparency published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently grounded it in the named Transparency & explainability Principle (para 38: be informed of AI-based decisions + right to request explanation).
- redress published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited para 55 (harms investigated and redressed via enforcement + remedial actions) and the Responsibility & accountability principle (paras 42-43).
- education published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited the named Education and Research Policy Area (para 101: provide adequate AI literacy education to the public).
- healthcare published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited the named Health and Social Well-being Policy Area (para 121: employ effective AI for health + the right to life; paras 122-123 safety/efficacy + bias oversight).
- employment published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited the named Economy and Labour Policy Area (para 116: assess and address AI's impact on labour markets).
- training_data published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited the named Data Policy area (para 71: data-governance strategies ensuring continual evaluation of training-data quality).
- environmental_impact_of_training published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited the named Environment and Ecosystems Policy Area (para 84: assess direct/indirect environmental impact incl. carbon footprint + energy consumption across the life cycle).
- international_coordination published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited the named Development and International Cooperation Policy Area (para 80: provide platforms for international cooperation on AI).
- development_rights_framing published governs → blind panel governs ✓ corroborated
Corroborated (3/3 governs). The panel independently cited the Development and International Cooperation Policy Area (para 79: AI-for-development bound to the values/principles, esp. for LMICs) + the Diversity & inclusiveness Principle (para 67).
- biometric_id published implicit → blind panel governs ⚑ diverged (adjacent) — flagged
Divergent by one step (flagged for review). The blind panel split 2-1 governs (para 26 no mass surveillance/social scoring + para 74 biometric data safeguards). The catalog kept implicit on reconciliation: these are general-principle hooks, not a dedicated biometric-ID operative section. The catalog's conservative call stands pending editor review.
- criminal_justice published implicit → blind panel governs ⚑ diverged (adjacent) — flagged
Divergent by one step (flagged for review). The blind panel read 3/3 governs (paras 62-63 name law enforcement + the judiciary as sensitive use cases requiring oversight). The catalog kept implicit: this is the general sensitive-use-case framing, not a dedicated criminal-justice Policy Area with topic-specific rules. The catalog's conservative call stands.
- ai_worker_displacement published implicit → blind panel governs ⚑ diverged (adjacent) — flagged
Divergent by one step (flagged for review). The blind panel read 3/3 governs (para 118: upskilling/reskilling + fair transition for at-risk workers). The catalog kept implicit: para 118 is a sub-provision of the Economy and Labour area already scored governs via 'employment', so scoring it governs again would double-count the area. The catalog's conservative call stands.
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
RATIFIED + PUBLISHED 2026-06-21 by the named editor (operator). Source URL resolves (unesco.org; full text cross-checked against the UNESCO PDF). 9 governs + 3 implicit + 11 silent (omitted). On ratification the verdicts were RECONCILED to the catalog convention — "governs" = an explicit operative topic-specific provision regardless of binding force (soft-law peers G7-Hiroshima/OECD/UN-Res/Bletchley/NIST-GenAI all carry governs cells). 9 cells with a dedicated named Policy Area or Principle + a verbatim para-anchored operative excerpt were upgraded implicit→governs: transparency (para 38), redress (para 55), education (para 101), healthcare (para 121), employment (para 116), training_data (para 71), environmental_impact_of_training (para 84), international_coordination (para 80), development_rights_framing (para 79). Confidence capped at medium (non-binding soft law). 3 kept implicit: biometric_id (general proportionality principle, no dedicated provision), criminal_justice (sensitive-use-case framing, paras 62-63, not a dedicated regime), ai_worker_displacement (para 118 sub-provision of the Economy & Labour area already scored via employment). An independent blind 3-analyst re-derivation (iter-432) corroborated the 9 governs (3/3 each) and flagged the 3 implicits as conservative; see /wiki/ai-curation.
Reviewer: PW contribute-instrument Workflow (§7.11) — research + web-verify (UNESCO primary source; OHCHR-hosted submission cross-check) → 23-topic classification → INDEPENDENT refute-by-default per non-silent cell.
✓ §7.11(a) citation + attestation✓ §7.12(b) operative provision excerpted
Primary source: TAKE IT DOWN Act, Pub. L. No. 119-12, 139 Stat. 55 (2025) (platform notice-and-removal at 47 U.S.C. § 223 / § 223a note (Communications Act of 1934 § 223), FTC-enforced under the FTC Act (15 U.S.C. § 57a); criminal provisions at 18 U.S.C. §§ 2252, 2256, 2264; the borrowed 'intimate visual depiction' definition is from 15 U.S.C. § 6851) · reviewed 2026-06-30
Governs verdicts — each carries its operative provision (§7.12(b)):
- deepfakes · high · Pub. L. 119-12 — criminalizes nonconsensual intimate 'digital forgeries' (AI deepfakes) of adults and minors and requires covered platforms to remove them within 48 hours; the statute names 'artificial intelligence' in its operative digital-forgery definition
paraphrase'Digital forgery' [is an intimate visual depiction] created through the use of software, machine learning, artificial intelligence, or any other computer-generated or technological means … indistinguishable from an authentic visual depiction.
Review verdict (the §7.12(c) adversarial classification trail — where the “supports” judgement lives)
Cleared on independent re-review. deepfakes=GOVERNS survived against explicit operative provisions — the statute names 'artificial intelligence' in its 'digital forgery' definition and imposes both criminal liability and a 48-hour platform takedown for nonconsensual intimate deepfakes (a rare deepfake cell grounded in named operative AI provisions rather than implication). redress=IMPLICIT survived and was NOT upgraded to governs: although the act provides two victim-remedy mechanisms (48-hour takedown + mandatory criminal restitution / forfeiture), they are narrow to the nonconsensual-intimate-image harm domain and there is no private right of action (FTC-exclusive enforcement), so it is incidental redress within a content-crime statute rather than a horizontal AI-redress regime. Three over-claim temptations were tested and rejected as silent: synthetic_content_provenance (the act mandates no watermarking/labeling — takedown only), transparency (the only disclosure is procedural notice of the removal process itself), and criminal_justice (the act creates criminal offenses but does not govern AI used within the criminal-justice system). AI-curated at reduced confidence; the named editor may confirm or correct.
Reviewer: PW autonomous adversarial classification review (§7.11) — refute-by-default vs Pub. L. 119-12, cross-corroborated across the FTC statute page, Latham & Watkins, Orrick, Skadden, and CRS LSB11314 (congress.gov / govinfo PDF bot-blocked, so verified via authoritative mirrors)