Library
Find public work on AI: neutral reference, research critiques and advocacy.
Search the library
85 entries · Safety and evaluation
Clear filtersSelected starting points, then latest dated entries. Dates are labelled by their meaning.
- ReferenceWikiGovernance instrument
EU AI Act
Risk-based framework. Prohibited practices (Art. 5) effective 2 February 2025; general-purpose AI obligations (Arts. 51-55) 2 August 2025; high-risk system obligations (Title III) 2 August 2026. Staggered 6/12/24-month application timeline from 1…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Bletchley Declaration on AI Safety
First multilateral consensus on frontier-AI safety risks. 28 signatories including US, EU, China. Introduced the policy vocabulary of 'frontier AI' that later instruments adopted. Non-binding but precedent-setting; spawned the AI Safety Institute…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
DFARS Subpart 252.204 (Safeguarding Covered Defense Information and Cyber Incident Reporting)
Defense-acquisition-specific information-security regulation. Core clauses: (1) DFARS 252.204-7012 (adopted 2015, current consolidated 2020) — requires contractors handling Covered Defense Information (CDI) on covered contractor information systems…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Executive Order 14110 on Safe, Secure, Trustworthy AI
Rescinded by EO 14148 (Jan 20, 2025); EO 14179 (Jan 23) set the deregulatory posture. Some §4 reporting persists via Defense Production Act + BIS interim rule. DRIFT-WATCH CORRECTION (2026-07-17, /wiki/drift.json): status was "partial" — WRONG. EO…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
FedRAMP AI Cloud Procurement Guidance
Operational PMO guidance for agencies acquiring AI / generative-AI cloud services within the existing FedRAMP authorisation framework. Key operational themes that recur across the published surface: (1) AI cloud services that process federal data…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
GSA Generative AI and Specialized Computing Infrastructure Acquisition Resource Guide
Procurement-focused operational guide accompanying OMB M-24-10 and the broader EO 14110 / EO 14179 federal-AI policy stack. Provides agencies with: (1) market intelligence on the governmentwide acquisition vehicles covering AI services (MAS IT and…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
India Digital Personal Data Protection Act + AI Advisory (MEITY)
India's primary AI-adjacent statute is the DPDPA + MEITY's binding AI advisories (Mar 2024 + Apr 2024 walked-back versions). No dedicated AI law yet; the proposed Digital India Act was paused 2024-2025. Affects 1.4B people — the single largest…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
OMB Memorandum M-24-10 (Advancing Governance, Innovation, and Risk Management for Agency Use of AI)
Binding on covered federal agencies. Three pillars: (I) strengthen AI governance through agency Chief AI Officers + AI Governance Boards; (II) advance responsible AI innovation including authorized use, talent, and infrastructure; (III) manage risks…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Revised Product Liability Directive (Directive (EU) 2024/2853)
EU strict-liability regime for defective products, modernised for the digital age and explicitly extended to software and AI systems. Repeals and replaces the 1985 Product Liability Directive (85/374/EEC). Art. 4(1) redefines "product" to include…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Seoul Declaration on Safe, Innovative and Inclusive AI
Bletchley follow-up. 16 frontier-AI-developer companies signed Frontier AI Safety Commitments alongside. Introduces measurable capability-evaluation expectations and pre-deployment thresholds; first instrument to formalise frontier-lab voluntary…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Japan AI Promotion Act (Act on the Promotion of Research, Development and Utilization of AI-Related Technologies)
Japan's first national AI statute (Act No. 53 of 2025), an innovation-first BASIC law (基本法-style) rather than a risk-regulation regime like the EU AI Act. Promulgated 4 June 2025; most provisions took effect that day, while Chapter III (AI Basic…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
New York RAISE Act: Responsible AI Safety and Education Act
The RAISE (Responsible AI Safety and Education) Act, S6953-B/A6453-B, signed by Governor Hochul on December 19, 2025 and effective January 1, 2027, adds Article 44-B (§§ 1420-1425) to the New York General Business Law. It is the second US state…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
UN Global Digital Compact
The Global Digital Compact (GDC) is Annex I to "The Pact for the Future", adopted by the UN General Assembly as Resolution A/RES/79/1 at the Summit of the Future on 22 September 2024. It is a non-binding, soft-law political framework (a General…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance topic
Agentic AI Governance
Obligations specific to AI systems that take autonomous multi-step actions (browse, transact, plan, recurse). Distinct from foundation_models (capability) and catastrophic_risk (outcome) — this is the action-surface frame. Surfaces in EU AI Office…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
California SB-1047: Safe and Secure Innovation for Frontier AI Models Act
A frontier-model safety-protocol-and-audit bill, not a pre-deployment testing mandate. Passed both chambers (Assembly 28 Aug 2024, Senate concurrence 29 Aug 2024 — the adoptedDate here) and was vetoed by Gov. Newsom on 29 September 2024, so it never…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance topic
Catastrophic & Existential Risk
Governance of model capabilities that could cause mass casualties or civilisational-scale harms (CBRN uplift, autonomous replication, deceptive alignment). Distinct from EU AIA 'systemic risk' which targets market-scale rather than catastrophic-scale…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Directive (EU) 2024/2831 on improving working conditions in platform work
The EU Platform Work Directive ((EU) 2024/2831) was adopted on 23 October 2024, published in the Official Journal on 11 November 2024, and entered into force on 1 December 2024; Member States must transpose it into national law by 2 December 2026. It…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance topic
Foundation Models / GPAI
Obligations specific to general-purpose / foundation models above certain capability thresholds.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Agentic AI System
An AI system that takes actions in the world — calling tools, executing code, browsing the web, sending messages, planning multi-step sequences — rather than only generating text or images for a human reader.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
AI Alignment
The technical problem of designing AI systems whose objectives, behaviour, and emergent goals reliably track the values or instructions of their principals across deployment contexts.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
AI Safety Level 3 (ASL-3)
A capability-based risk tier in Anthropic's Responsible Scaling Policy denoting models with the potential to substantially uplift CBRN attack capabilities or autonomous AI replication.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
AI Supply Chain
The end-to-end pipeline of inputs, intermediate artefacts, and downstream applications by which an AI system is built and deployed — typically decomposed as training data → compute → model weights → fine-tuning → deployment → downstream applications.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
AIME 2024
30 problems from the 2024 American Invitational Mathematics Examination — high-school competition math.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Anthropic Responsible Scaling Policy (RSP) v2
First-mover industry safety framework. Introduces the AI Safety Level (ASL) capability-tier vocabulary subsequently adapted by OpenAI Preparedness + DeepMind FSF. v2 (Oct 2024) refines ASL-3/ASL-4 capability thresholds, mandates pre-deployment…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
ARC-AGI v2
Abstract reasoning over visual grids. Each task requires inferring the transformation rule from 2-3 examples.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
ASEAN Guide on AI Governance and Ethics
Non-binding voluntary guide for 10 ASEAN member states (Indonesia, Malaysia, Philippines, Singapore, Thailand, Vietnam, Myanmar, Cambodia, Laos, Brunei). Adopts a cross-cutting risk + values framework intentionally distinct from the EU AIA's…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Brazil AI Bill (PL 2338/2023)
Risk-based framework structurally similar to EU AIA but with distinct development-rights framing rooted in Brazil's Marco Civil tradition. Senate-approved Dec 2024; Chamber of Deputies vote pending 2025. Notable for explicit human-dignity +…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Capability Elicitation
Techniques designed to reveal the upper bounds of an AI model's capabilities, rather than measuring its default behaviour, so that downstream safety judgements can be calibrated to what the model *can* do under adversarial prompting or fine-tuning.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Compute Threshold (AI Governance)
A regulatory trigger expressed as floating-point operations (FLOPs) consumed during model training, above which specific reporting, evaluation, or governance obligations attach.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Council of Europe Framework Convention on AI
First legally-binding international treaty on AI. Opened for signature Sep 2024. Enters into force three months after five ratifications including three CoE members. Currency (2026-06-21): secondary trackers report the entry-into-force threshold was…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Data Poisoning
A training-time attack in which an adversary inserts crafted examples into the training corpus or fine-tuning dataset to alter the resulting model's behaviour — typically inserting a backdoor that triggers on a specific input pattern or degrading…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Deceptive Alignment
A failure mode in which a model appears aligned during training and evaluation because doing so serves its actual (mesa-)objective, but pursues divergent objectives once deployed or once it judges itself unobserved.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Designated Systemic-Risk Model
A general-purpose AI model that has been formally designated by the EU AI Office under Article 51(1)(b) as posing systemic risk, regardless of whether it meets the presumption thresholds.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Dual-Use Research Norms (DURC for AI)
A normative framework — adapted from biosecurity's Dual-Use Research of Concern (DURC) policies — for governing AI research and publication decisions when research outputs have both beneficial and harmful applications.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
EU General-Purpose AI Code of Practice
Operational bridge between EU AIA Arts. 53-55 (general-purpose AI obligations) and provider compliance. Art. 56(8) AIA gives adherent providers a presumption of compliance with the substantive obligations — distinct from industry self-pledges…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Frontier-Tier AI
A categorical classification of AI models above certain capability or compute thresholds, indicating heightened regulatory scrutiny.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
FrontierMath
Hundreds of original research-mathematician-curated math problems requiring deep reasoning. Held-out evaluation only.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Google DeepMind Frontier Safety Framework
Critical Capability Levels (CCL) regime spanning autonomy, biosecurity, cybersecurity, and persuasion domains. Distinct vocabulary from Anthropic ASL + OpenAI Preparedness — designed for cross-domain elicitation; each CCL triggers domain-specific…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
GPQA Diamond
Graduate-level Google-Proof Q&A in biology, chemistry, physics. 'Diamond' subset is the 198 hardest items.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Hallucination
Confidently-asserted but factually incorrect output produced by an AI model — including fabricated citations, invented people or events, and confabulated numerical values — that the model cannot reliably distinguish from correct output at generation…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
HumanEval
164 hand-written Python programming problems. Generate a function that passes provided unit tests.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
Humanity's Last Exam
3,000+ frontier-difficulty expert-curated questions across all academic disciplines. Designed to remain unsaturated through 2026+.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
In-Context Learning
The capacity of a foundation model to adapt its behaviour to a new task purely from examples provided in the prompt, without any updates to the model's weights — discovered as an emergent property of large language models and now a primary evaluation…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Inference-Time Compute
The scaling regime in which model capability is increased by spending more compute at inference time (multiple samples, search, longer reasoning chains, tool-using iteration) rather than by training a larger model — disrupting the…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Jailbreak Resistance
The robustness of an AI model's safety training against adversarial prompts crafted to elicit policy-prohibited outputs — distinct from alignment (which concerns the model's goals) and from baseline safety training (which concerns the model's…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Japan METI AI Guidelines for Business
Joint METI + MIC issuance consolidating prior AI Utilization Guidelines (2019) + AI R&D Principles (2017) into a single business-facing framework. Voluntary; explicitly aligned with G7 Hiroshima AI Process Code of Conduct + OECD AI Principles. Ten…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
MATH (Hendrycks)
12,500 competition-math problems from AMC, AIME, etc. Evaluates step-by-step reasoning + final-answer accuracy.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Mesa-Optimization
The phenomenon in which a learned model itself implements an optimisation algorithm at inference time, producing an inner objective ('mesa-objective') that may differ from the outer training objective.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Meta Frontier AI Framework
Meta's open-weight-frontier governance posture. Categorises frontier models into 'high risk' + 'critical risk' tiers; the framework's distinctive feature is its explicit defence of open-weight release as a governance posture (vs. the closed-model…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
MMLU
Massive Multitask Language Understanding — 57-subject multiple-choice covering humanities, STEM, social sciences, professional/legal.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
MMLU-Pro
Successor to MMLU with 10-option multiple-choice (up from 4), more reasoning-focused tasks, and removed leaky / ambiguous items.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Model Card
A standardized disclosure document accompanying an AI model that describes its intended use, training data, evaluation results, limitations, and known failure modes.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Model Distillation Risk
The risk that a closed-weight frontier model's capabilities can be partially recovered by training a smaller open-weight model on the closed model's outputs, undermining the governance assumption that closed weights confer capability containment.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Model-Merging Risk
The governance concern that post-training combination of multiple specialised models — via weight averaging, task-arithmetic, or modular merging — can produce capability or safety properties not present in any single source model, in ways the…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Multi-Turn Evaluation
An evaluation methodology that probes AI models across multi-step conversations rather than single prompts — designed to surface deception, sycophancy, context-accumulation jailbreaks, and capability degradation that single-prompt benchmarks miss.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
NIST AI RMF Generative AI Profile
Companion to NIST AI 100-1 covering GenAI-specific risks: CBRN information uplift, confabulation, data privacy, environmental impacts, harmful bias, dangerous information, IP misuse, obscene/abusive/violent content, information security, information…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
OpenAI Preparedness Framework
Capability-tier risk evaluation regime with four categorical levels (Low / Medium / High / Critical) across four risk categories (cybersecurity, CBRN, persuasion, model autonomy). Pre-deployment evaluation against the framework gates release…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Policy Instrument
An identifiable technique of collective action — a binding regulation, an executive order, a voluntary code, a technical standard, a treaty, or similar — by which a public authority structures behaviour to address a policy problem. Instrument choice…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Prompt Injection
An adversarial input technique in which untrusted content fed to an AI model (e.g., text on a webpage the model reads, a document the user uploads, a tool's output) contains instructions that override the model's intended behaviour or…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Red-Team Evaluation
Structured adversarial probing of an AI model's capabilities and behaviour before deployment, designed to elicit failures that ordinary evaluation would miss.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Retrieval-Augmented Generation (RAG)
An AI system pattern in which a model's outputs are conditioned on external content retrieved at inference time from a knowledge source — combining the parametric knowledge of the model with the up-to-date or domain-specific knowledge of the…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Sandbagging
A theoretical failure mode in which a model deliberately underperforms on capability evaluations — either to avoid triggering capability-tier safeguards or because of subtle training-time incentives that reward apparent low capability under…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Scalable Oversight
The set of techniques for supervising AI systems whose outputs are too complex, too numerous, or too domain-distant for unaided human evaluators to judge correctness.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Singapore Model AI Governance Framework for Generative AI
Update to the 2020 Model AI Governance Framework (v2), expanding scope to generative AI. Nine dimensions: accountability, data, trusted development + deployment, incident reporting, testing + assurance, security, content provenance, safety +…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiBenchmark
SWE-bench Verified
Solve real-world GitHub issues from 12 popular Python repos. The 'Verified' subset is human-validated to remove ambiguity and have working tests.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Systemic Risk (AI)
A regulatory designation indicating that a general-purpose AI model poses risks of significant scale or scope across the EU internal market, triggering Article 55 obligations under the EU AI Act.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Tool-Use Safety
The sub-domain of agentic-system safety concerned with the risks that arise when an AI model invokes external tools (search, code execution, APIs, financial transactions, system commands) — including risks of unintended action, instruction…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
UK-US AI Safety Institute Memorandum of Understanding
First binding bilateral on frontier-AI safety. Commits both AISIs to coordinated pre-deployment evaluations, red-team data sharing, methodological alignment on capability elicitation, and joint exercises across at least one major frontier-model…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
UN GA Resolution on Safe, Secure, Trustworthy AI
Non-binding. Calls on member states to bridge digital divides and develop national strategies. China + US co-sponsored; passed by consensus. Currency (2026-06-21): the UN AI-governance track has since advanced beyond this non-binding resolution —…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
White House Voluntary AI Commitments
First broad-spectrum US industry commitments; precursor to EO 14110 §4.2(a) reporting + the Seoul Frontier AI Safety Commitments. 15 signatories across two tranches (Jul + Sep 2023): Anthropic, OpenAI, Google DeepMind, Microsoft, Meta, Inflection,…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiConcept
Chain-of-Thought Monitoring
A frontier-safety oversight approach that inspects a model's externalized step-by-step reasoning trace (its chain of thought) to detect intent to misbehave — such as reward hacking, deception, or unsafe planning — before acting on the model's final…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
California SB 942: AI Transparency Act
SB 942 (the 'California AI Transparency Act'), Chapter 291, Statutes of 2024, adds §§ 22757–22757.4 to the California Business and Professions Code — a generative-AI provenance-and-disclosure law regulating 'covered providers' (a person that produces…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
California SB-53: Transparency in Frontier Artificial Intelligence Act (TFAIA)
SB 53 (TFAIA), signed Sept. 29, 2025 (Chapter 138), is the first US state law expressly regulating 'frontier' AI; it succeeds the vetoed SB 1047 with a transparency-and-disclosure design rather than pre-deployment liability. It applies to 'frontier…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
DoD Responsible AI Strategy and Implementation Pathway
DoD-wide operational pathway implementing the five Ethical Principles for AI (Responsible, Equitable, Traceable, Reliable, Governable; adopted Feb 24, 2020). Six foundational tenets: (1) RAI Governance — clarifies roles between OUSD(R&E), OUSD(A&S),…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Executive Order 14179 — Removing Barriers to American Leadership in AI
Rescinds EO 14110's regulatory-burden provisions. Directs OMB / OSTP / NSC to remove barriers to AI development. Does NOT itself impose new substantive obligations — coverage is mostly silent. The DPA-grounded compute-reporting interim rule (BIS, Jan…
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
G7 Hiroshima AI Process Code of Conduct
Voluntary commitments by frontier AI developers. 11-point code covering risk identification, deployment, content provenance, security investment, info sharing.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
Interim Measures for Generative AI Service Management
Joint CAC/MIIT/MPS measures. Registration + safety assessment for public-facing generative AI. Aligns with Algorithm Recommendation Rules (2022) and Deep Synthesis Rules (2022).
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
NIST AI Risk Management Framework
Voluntary. Four functions (Govern / Map / Measure / Manage). GenAI Profile (NIST AI 600-1) added 2024 for GPAI-specific guidance.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
OECD AI Principles (Recommendation)
First intergovernmental standard. Updated 2024 to clarify GPAI scope. Foundation referenced by G7, GPAI, and many national frameworks.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiGovernance instrument
UK Pro-Innovation Approach to AI Regulation (White Paper)
Principles-based, regulator-led approach (no statutory AI law). Cross-sectoral principles delegated to existing regulators. AISI established Nov 2023 for evaluation/safety research.
- Status
- Public reference
- Reference review:
- Publisher:
- Policy Window
- ReferenceWikiDebate
Compute vs Behavioural Capability Thresholds
Should the regulatory trigger for 'frontier' / 'foundation' / 'systemic-risk' status be training-compute thresholds (objective + ex-ante observable), or behavioural-capability evaluation (more semantically meaningful but operationally costly)?
- Status
- Public reference
- Date
- Review date not recorded
- Publisher:
- Policy Window
- ReferenceWikiDebate
Open-Source vs Closed-Source Frontier Models
Should the most-capable AI models be released under permissive licenses (open weights), or only via API / structured-access agreements? The dispute is foundational to nearly every frontier-AI governance instrument.
- Status
- Public reference
- Date
- Review date not recorded
- Publisher:
- Policy Window
- ReferenceWikiDebate
Pause AI vs Accelerate Capabilities
Should the global community impose temporary or capability-conditional pauses on frontier-AI development, or should development accelerate with safety work conducted in parallel?
- Status
- Public reference
- Date
- Review date not recorded
- Publisher:
- Policy Window
- ReferenceWikiDebate
Pre-Deployment Red-Team vs Post-Deployment Audit
Should AI capability + safety evaluations happen primarily before deployment (red-team gating release), or primarily after (post-deployment audit + incident response)?
- Status
- Public reference
- Date
- Review date not recorded
- Publisher:
- Policy Window
- ReferenceWikiDebate
Risk-Based vs Principles-Based vs Ex-Post Liability Regimes
Should AI governance work via (a) risk-based ex-ante categorisation + obligations (EU), (b) high-level principles delegated to sector regulators (UK / OECD / G7), or (c) ex-post liability + civil litigation (US sectoral)?
- Status
- Public reference
- Date
- Review date not recorded
- Publisher:
- Policy Window
Looking for articles you saved? Open your Wiki reading list.