Cross-corpus research synthesis
Compute-Threshold Reporting
Mandatory reporting based on training-compute or capability thresholds.
Synthesised deterministically from 34 articles that engage this theme. Empirical consensus: contested · contested: Are compute thresholds (10²⁵ FLOPs EU, 10²⁶ FLOPs US) a defensible proxy for governance-relevant capability, given algorithmic-efficiency improvements? Field is split.. Full theme article: /wiki/compute-reporting. Machine-readable: /wiki/synthesis.json.
Cross-jurisdiction stances (5 govern, 16 engage)
| Instrument | Verdict | Provision excerpt / citation |
|---|---|---|
| EU AI Act | governs | a general-purpose AI model shall be presumed to have high impact capabilities … when the cumulative amount of computation used for its training measured in floating point operations is greater than 10^25. (paraphrase) Art. 51(2) + Annex XIII (10²⁵ FLOP presumption) |
| Executive Order 14110 on Safe, Secure, Trustworthy AI | governs | §4.2(a)(i) — 10²⁶ FLOP threshold |
| Bletchley Declaration on AI Safety | implicit | Declaration §6 calls for capability evaluation but does not specify compute thresholds |
| Seoul Declaration on Safe, Innovative and Inclusive AI | implicit | Safety Commitments invoke capability thresholds; compute is one proxy |
| California SB-1047: Safe and Secure Innovation for Frontier AI Models Act | governs | Cal. SB-1047 §22603(b) — annual reporting of training compute + safety determination |
| Anthropic Responsible Scaling Policy (RSP) v2 | implicit | RSP v2 capability evaluations triggered by capability rather than pure compute; compute is one signal |
| OpenAI Preparedness Framework | implicit | Capability-tier evaluations are the primary trigger; compute is a coincident signal |
| White House Voluntary AI Commitments | implicit | Self-reporting through commitments framework; binding compute thresholds came via EO 14110 §4.2(a) |
| OMB Memorandum M-24-10 (Advancing Governance, Innovation, and Risk Management for Agency Use of AI) | governs | Agencies must report to OMB and, as appropriate, publicly release aggregate metrics about their AI use cases that are determined to be safety-impacting or rights-impacting. (paraphrase) §3(a)(iv)–(v) annual public AI use-case inventory + quarterly AI procurement reporting to OMB |
| GSA Generative AI and Specialized Computing Infrastructure Acquisition Resource Guide | governs | Faithful summary: the guide routes AI acquisitions through existing governmentwide vehicles (MAS IT and Best-in-Class GWACs), noting there is no generative-AI-only vehicle; it does not enumerate new AI-specific Special Item Numbers. (paraphrase) Guide routes AI acquisitions through existing governmentwide vehicles (MAS IT / Best-in-Class GWACs) rather than a dedicated generative-AI vehicle or new AI-specific SINs |
| DoD Responsible AI Strategy and Implementation Pathway | implicit | Tenet 1 (RAI Governance) + Tenet 3 (Acquisition Lifecycle) — clarifies CDAO + OUSD(A&S) roles in AI procurement oversight; tracking + reporting emerge through standard DoD acquisition reporting channels |
| FedRAMP AI Cloud Procurement Guidance | implicit | FedRAMP authorisation enables ATO; agency-AI-use disclosure flows through OMB M-24-10 inventory + quarterly procurement reporting rather than through FedRAMP itself |
| DFARS Subpart 252.204 (Safeguarding Covered Defense Information and Cyber Incident Reporting) | implicit | Cyber-incident reporting under 252.204-7012(c) — 72-hour DoD notification covers AI-system compromise events including model-weight theft + prompt-injection-based credential exposure; broader AI-use disclosure flows through M-24-10 not DFARS |
| California SB-53: Transparency in Frontier Artificial Intelligence Act (TFAIA) | implicit | Bus. & Prof. Code § 22757.11 uses a 10^26 FLOP compute threshold to SCOPE the regulated class + § 22757.12 ties disclosure to compute-defined frontier models; no standalone compute-figure reporting mandate to a regulator |
| New York RAISE Act: Responsible AI Safety and Education Act | implicit | N.Y. Gen. Bus. Law § 1420(6),(9) — the frontier-model / large-developer compute figures SCOPE the regulated class; no standalone compute-figure reporting duty to a regulator. (The Mar. 27, 2026 chapter amendment revised the large-developer threshold to align more closely with California's criteria; the verdict — coverage-scoping, not a reporting duty — is unchanged by the specific figure.) |
| Japan AI Promotion Act (Act on the Promotion of Research, Development and Utilization of AI-Related Technologies) | implicit | ... facilities and equipment relating to large-scale information processing ... the State shall take measures to develop, improve, and promote the shared use of such facilities ... (paraphrase) Act No. 53 of 2025, Art. 12 |
Evidence convergence
Sources the corpus cites for this theme across multiple articles — a scientometric consensus signal computed from inline prose citations (the more articles independently cite a source, the more load-bearing it is for this theme). 41 sources are cited by ≥2 articles.
- 26×Defending Compute Thresholds Against Legal Loopholes — cited by 26 articles
- 23×An interdisciplinary account of the terminological choices by EU policymakers ahead of the final agreement on the AI Act: AI system, general purpose AI system, foundation model, and generative AI — cited by 23 articles
- 17×The EU model of AI governance: regulating artificial intelligence through law and policy — cited by 17 articles
- 16×Training Compute Thresholds: Features and Functions in AI Regulation — cited by 16 articles
- 16×Computing Power and the Governance of Artificial Intelligence — cited by 16 articles
- 9×Evaluating Frontier Models for Dangerous Capabilities — cited by 9 articles
- 8×Two types of AI existential risk: decisive and accumulative — cited by 8 articles
- 8×GPTs are GPTs: Labor market impact potential of LLMs — cited by 8 articles
- 8×Generative AI and data protection — cited by 8 articles
- 8×Artificial intelligence and synthetic biology: biosecurity risks, dual-use concerns, and governance pathways — cited by 8 articles
- 7×Verification methods for international AI agreements — cited by 7 articles
- 6×Compute North vs. Compute South: The Uneven Possibilities of Compute-based AI Governance Around the Globe — cited by 6 articles
- 5×Frontier AI Regulation: Managing Emerging Risks to Public Safety — cited by 5 articles
- 5×Open Problems in Technical AI Governance — cited by 5 articles
- 5×Governing Through the Cloud: The Intermediary Role of Compute Providers in AI Regulation — cited by 5 articles
- 5×Identifying Algorithmic Decision Subjects' Needs for Meaningful Contestability — cited by 5 articles
- 5×Governing AI Agents — cited by 5 articles
- 5×International Agreements on AI Safety: Review and Recommendations for a Conditional AI Safety Treaty — cited by 5 articles
- 5×Digital Disintegration: Techno-Blocs and Strategic Sovereignty in the AI Era — cited by 5 articles
- 4×arxiv:2504.18236 — cited by 4 articles
Concepts in play
Frontier-Tier AISystemic Risk (AI)Designated Systemic-Risk ModelCompute Threshold (AI Governance)Red-Team EvaluationAI AlignmentDeceptive AlignmentMesa-OptimizationCapability ElicitationPolicy InstrumentAI Supply ChainMulti-Turn EvaluationModel Distillation RiskInference-Time ComputeSandbaggingIn-Context LearningHardware-Enabled Governance Mechanisms