Scope and obligations

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 capability evaluations, and commits to a Frontier Red Team. Seoul Frontier AI Safety Commitments signatory; cited by name in EU AI Office GPAI Code of Practice drafts. NOTE (iter-314): the RSP is a versioned-evolving artefact; this row pins v2 (Oct 2024) as the load-bearing reference, but Anthropic publishes incremental updates on the policy page. Citers tracking specific ASL-4 threshold language should confirm against the current version on anthropic.com — the catalog re-pins on the next Coverage Games event.

Anthropic Responsible Scaling Policy (RSP) v2 addresses 4 contested AI-governance topics explicitly, 6 via general principles,.

Topics governed

• governs
  Foundation Models / GPAI— RSP v2 §2 — ASL framework applies to frontier model releases
• implicit
  Compute-Threshold Reporting— RSP v2 capability evaluations triggered by capability rather than pure compute; compute is one signal
• governs
  Transparency Obligations— RSP v2 §5 — public publication of safety determinations + capability eval methodology
• governs
  Catastrophic & Existential Risk— RSP v2 §3 — ASL-3 / ASL-4 capability thresholds explicitly target CBRN uplift + autonomous-replication
• implicit
  International Coordination— Seoul Frontier AI Safety Commitments signatory; coordinates with US + UK AISIs on capability evaluation
• governs
  Agentic AI Governance— RSP v2 — ASL thresholds include 'autonomous AI replication' + agentic capability evaluations
• implicit
  Open-Weight Frontier Release— RSP applies to Anthropic's models which are closed-weight; framework does not address third-party open release
• implicit
  Synthetic Content Provenance— Deployment-stage controls would include content provenance where capability tier requires
• implicit
  AI in Elections— Anthropic's election-integrity acceptable-use policies apply to deployments; not in RSP itself
• implicit
  Compute + Model-Weight Export Controls— ASL-3+ tiers include model-weight access controls (recipient-restriction analog)

Cross-jurisdiction comparison

How peer instruments treat the topics Anthropic Responsible Scaling Policy (RSP) v2 governs.

°= industry self-imposed voluntary framework. Comparing a voluntary code's "governs" tint with a binding regulation's "governs" tint flattens the legal-force distinction; use the instrument-page banner for the operative status of each.

How to cite this article

Policy Window. (2024). Anthropic Responsible Scaling Policy (RSP) v2 [Wiki article — Instrument]. https://policywindow.org/wiki/anthropic-rsp

Policy Window. 2024. "Anthropic Responsible Scaling Policy (RSP) v2." Wiki article (Instrument). https://policywindow.org/wiki/anthropic-rsp.

@misc{policywindow-anthropic-rsp,
  title  = {Anthropic Responsible Scaling Policy (RSP) v2},
  author = {Policy Window},
  year   = {2024},
  howpublished = {Anthropic Responsible Scaling Policy v2 (Oct 2024)},
  url    = {https://policywindow.org/wiki/anthropic-rsp},
  note   = {Primary source: https://www.anthropic.com/news/announcing-our-updated-responsible-scaling-policy}
}

Related instruments

• EU AI Act · EU
• Executive Order 14110 on Safe, Secure, Trustworthy AI · US
• G7 Hiroshima AI Process Code of Conduct · G7
• NIST AI Risk Management Framework · US
• Bletchley Declaration on AI Safety · global
• Seoul Declaration on Safe, Innovative and Inclusive AI · global
• NIST AI RMF Generative AI Profile · US
• California SB-1047: Safe and Secure Innovation for Frontier AI Models Act · US
• Brazil AI Bill (PL 2338/2023) · BR
• OpenAI Preparedness Framework · US
• Google DeepMind Frontier Safety Framework · US
• Meta Frontier AI Framework · US
• White House Voluntary AI Commitments · US
• Singapore Model AI Governance Framework for Generative AI · SG
• Japan METI AI Guidelines for Business · JP
• EU General-Purpose AI Code of Practice · EU

References

1. Anthropic Responsible Scaling Policy v2 (Oct 2024)
2. RSP v2 §2 — ASL framework applies to frontier model releases
3. RSP v2 capability evaluations triggered by capability rather than pure compute; compute is one signal
4. RSP v2 §5 — public publication of safety determinations + capability eval methodology
5. RSP v2 §3 — ASL-3 / ASL-4 capability thresholds explicitly target CBRN uplift + autonomous-replication
6. Seoul Frontier AI Safety Commitments signatory; coordinates with US + UK AISIs on capability evaluation
7. RSP v2 — ASL thresholds include 'autonomous AI replication' + agentic capability evaluations
8. RSP applies to Anthropic's models which are closed-weight; framework does not address third-party open release
9. Deployment-stage controls would include content provenance where capability tier requires
10. Anthropic's election-integrity acceptable-use policies apply to deployments; not in RSP itself
11. ASL-3+ tiers include model-weight access controls (recipient-restriction analog)

Scope and obligations

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 decisions; Safety Advisory Group + board-level Safety & Security Committee govern threshold determinations. Seoul Frontier AI Safety Commitments signatory. NOTE (iter-314): the Preparedness Framework is a versioned-evolving artefact; this row pins the originally-published Dec 2023 version, but OpenAI publishes updates on the safety/preparedness page. Citers tracking specific risk-category language or threshold definitions should confirm against the current published version — the catalog re-pins on the next Coverage Games event.

OpenAI Preparedness Framework addresses 3 contested AI-governance topics explicitly, 5 via general principles,.

Topics governed

• governs
  Foundation Models / GPAI— Preparedness Framework §1-2 — applies to all OpenAI frontier-model releases
• implicit
  Compute-Threshold Reporting— Capability-tier evaluations are the primary trigger; compute is a coincident signal
• implicit
  Transparency Obligations— Public Preparedness Reports + Safety Advisory Group decisions; full evaluation methodology partially disclosed
• governs
  Catastrophic & Existential Risk— Preparedness Framework risk-tier matrix — Critical tier explicitly targets CBRN, cyber, persuasion, autonomy
• implicit
  International Coordination— Seoul Frontier AI Safety Commitments signatory; pre-deployment evaluation sharing with US + UK AISIs
• governs
  Agentic AI Governance— Preparedness Framework — Model Autonomy is one of four named risk categories
• implicit
  Open-Weight Frontier Release— Framework applies to OpenAI deployments (closed-weight); does not address third-party open release
• implicit
  AI in Elections— Persuasion is one of four named risk categories; election-AI is the canonical persuasion case

Cross-jurisdiction comparison

How peer instruments treat the topics OpenAI Preparedness Framework governs.

°= industry self-imposed voluntary framework. Comparing a voluntary code's "governs" tint with a binding regulation's "governs" tint flattens the legal-force distinction; use the instrument-page banner for the operative status of each.

How to cite this article

Policy Window. (2023). OpenAI Preparedness Framework [Wiki article — Instrument]. https://policywindow.org/wiki/openai-preparedness

Policy Window. 2023. "OpenAI Preparedness Framework." Wiki article (Instrument). https://policywindow.org/wiki/openai-preparedness.

@misc{policywindow-openai-preparedness,
  title  = {OpenAI Preparedness Framework},
  author = {Policy Window},
  year   = {2023},
  howpublished = {OpenAI Preparedness Framework (Dec 2023)},
  url    = {https://policywindow.org/wiki/openai-preparedness},
  note   = {Primary source: https://openai.com/safety/preparedness}
}

Related instruments

• Executive Order 14110 on Safe, Secure, Trustworthy AI · US
• G7 Hiroshima AI Process Code of Conduct · G7
• Bletchley Declaration on AI Safety · global
• Seoul Declaration on Safe, Innovative and Inclusive AI · global
• NIST AI RMF Generative AI Profile · US
• California SB-1047: Safe and Secure Innovation for Frontier AI Models Act · US
• Brazil AI Bill (PL 2338/2023) · BR
• Anthropic Responsible Scaling Policy (RSP) v2 · US
• Google DeepMind Frontier Safety Framework · US
• Meta Frontier AI Framework · US
• EU General-Purpose AI Code of Practice · EU

References

1. OpenAI Preparedness Framework (Dec 2023)
2. Preparedness Framework §1-2 — applies to all OpenAI frontier-model releases
3. Capability-tier evaluations are the primary trigger; compute is a coincident signal
4. Public Preparedness Reports + Safety Advisory Group decisions; full evaluation methodology partially disclosed
5. Preparedness Framework risk-tier matrix — Critical tier explicitly targets CBRN, cyber, persuasion, autonomy
6. Seoul Frontier AI Safety Commitments signatory; pre-deployment evaluation sharing with US + UK AISIs
7. Preparedness Framework — Model Autonomy is one of four named risk categories
8. Framework applies to OpenAI deployments (closed-weight); does not address third-party open release
9. Persuasion is one of four named risk categories; election-AI is the canonical persuasion case