| Multi-lab | EU General-Purpose AI Code of Practice published | governance | 2024-08-01 | 2025-07-10 | Overdue | European Commission ↗ |
| Multi-lab | EU AI Act general-purpose AI obligations begin to apply | governance | 2024-08-01 | 2025-08-02 | Overdue | EU AI Act ↗ |
| OpenAI | Review the Preparedness Framework at least once a year | governance | 2025-04-15 | 2026-04-15 | Overdue | OpenAI ↗ |
| Anthropic | Publish a Risk Report every three to six months | transparency | 2026-02-24 | 2026-08-24 | Upcoming | Anthropic ↗ |
| Anthropic | Annual third-party review of Responsible Scaling Policy compliance | governance | 2026-02-24 | 2027-02-24 | Upcoming | Anthropic ↗ |
| Multi-lab | EU AI Act obligations apply to GPAI models already on the market | governance | 2024-08-01 | 2027-08-02 | Upcoming | EU AI Act ↗ |
| Multi-lab | EU AI Act high-risk (Annex III) obligations apply | governance | 2024-08-01 | 2027-12-02 | Upcoming | Hogan Lovells ↗ |
| Anthropic | Interpretability that can "reliably detect most model problems" by 2027 | evaluations | 2025-04-24 | 2027-12-31 | Upcoming | Dario Amodei (Anthropic) ↗ |
| Multi-lab | EU AI Act high-risk (Annex I) obligations apply | governance | 2024-08-01 | 2028-08-02 | Upcoming | Hogan Lovells ↗ |
| Multi-lab | EU AI Act first Commission evaluation and review | governance | 2024-08-01 | 2028-08-02 | Upcoming | EU AI Act ↗ |
| Anthropic | Define ASL-4 safeguards before reaching ASL-3 | safety-framework | 2023-09-19 | trigger: before any model reaches ASL-3 | Pending | Anthropic ↗ |
| Google DeepMind | Evaluate models at a compute / fine-tuning cadence | evaluations | 2024-05-17 | trigger: every 6x effective compute / 3 months fine-tuning (FSF v1.0) | Pending | Google DeepMind ↗ |
| xAI | Publish a frontier safety framework by the Paris AI Action Summit | safety-framework | 2024-05-21 | 2025-02-10 | Missed | The Midas Project ↗ |
| Mistral | Publish a frontier safety framework by the Paris AI Action Summit | safety-framework | 2024-05-21 | 2025-02-10 | Missed | METR ↗ |
| Multi-lab | Protect unreleased model weights (2023 voluntary commitments) | security | 2023-07-21 | trigger: safeguard unreleased frontier model weights (cybersecurity + insider threat) | Missed | Wang, Huang, Klyman & Bommasani (AIES 2025) ↗ |
| OpenAI | Dedicate 20% of compute to superalignment over four years | compute-pledge | 2023-07-05 | trigger: over four years from 2023-07 | Missed | Fortune ↗ |
| xAI | Publish an updated policy within three months | safety-framework | 2025-02-20 | 2025-05-10 | Partial | TechCrunch ↗ |
| Multi-lab | Publish a frontier safety framework before the Paris AI Action Summit | safety-framework | 2024-05-21 | 2025-02-10 | Partial | GOV.UK ↗ |
| Anthropic | Capability re-assessment cadence in the Responsible Scaling Policy | evaluations | 2023-09-19 | trigger: regular re-assessment cadence per the RSP | Partial | Anthropic ↗ |
| Multi-lab | Publicly report model capabilities and limitations (2023 voluntary commitments) | transparency | 2023-07-21 | trigger: publish a capabilities/limitations report with each significant model release | Partial | Wang, Huang, Klyman & Bommasani (AIES 2025) ↗ |
| Multi-lab | Incentivize third-party vulnerability reporting (2023 voluntary commitments) | security | 2023-07-21 | trigger: run/expand bounty mechanisms covering in-scope AI systems | Partial | Wang, Huang, Klyman & Bommasani (AIES 2025) ↗ |
| Multi-lab | Develop provenance or watermarking for AI-generated content (2023 voluntary commitments) | transparency | 2023-07-21 | trigger: develop provenance/watermarking for in-scope AI-generated audio-visual content | Partial | The White House (archive) ↗ |
| Multi-lab | Provide the UK AI Safety Institute pre-deployment model access | access | 2023-11-02 | trigger: pre-deployment access for state safety institutes | Partial | Center for AI Safety ↗ |
| Anthropic | Long-Term Benefit Trust to elect a majority of the board | governance | 2023-09-19 | 2027-09-19 | Met | Anthropic ↗ |
| Anthropic | Publish Responsible Scaling Policy v3.0 | safety-framework | 2026-02-24 | 2026-02-24 | Met | Anthropic ↗ |
| Google DeepMind | Implement the Frontier Safety Framework by early 2025 | safety-framework | 2024-05-17 | 2025-03-01 | Met | Google DeepMind ↗ |
| Amazon | Publish a frontier safety framework by the Paris AI Action Summit | safety-framework | 2024-05-21 | 2025-02-10 | Met | Amazon Science ↗ |
| Microsoft | Publish a Frontier Governance Framework (v1) | safety-framework | 2024-05-21 | 2025-02-10 | Met | Microsoft ↗ |
| Meta | Publish a frontier safety framework by the Paris AI Action Summit | safety-framework | 2024-05-21 | 2025-02-10 | Met | Meta ↗ |
| Multi-lab | Sign US AI Safety Institute access agreements with OpenAI and Anthropic | access | 2024-08-29 | 2024-08-29 | Met | NIST ↗ |
| OpenAI | Deliver Safety and Security Committee recommendations within 90 days | governance | 2024-05-28 | 2024-08-26 | Met | OpenAI ↗ |
| Anthropic | Publish a Responsible Scaling Policy (v1.0) | safety-framework | 2023-09-19 | 2023-09-19 | Met | Anthropic ↗ |
| Anthropic | Publish sabotage risk reports for future frontier models | transparency | 2025-11-24 | trigger: for each future frontier model clearly exceeding Opus 4.5 | Met | Anthropic ↗ |
| Anthropic | Apply ASL-3 safeguards when a model may reach the ASL-3 threshold | security | 2023-09-19 | trigger: when a model may reach the ASL-3 capability threshold | Met | Anthropic ↗ |
| Multi-lab | Internal and external security testing before model release (2023 voluntary commitments) | security | 2023-07-21 | trigger: before each model release | Met | The White House (archive) ↗ |