Examples
Examples from published documents
- Anthropic describes its Responsible Scaling Policy as "our voluntary framework for managing catastrophic risks from advanced AI systems" (Anthropic RSP v3.4, Introduction, p.3)
- "The Frontier Safety Framework will be reviewed at least once a year" (Google DeepMind Frontier Safety Framework v3.1, s.5.1, p.17)
- A large frontier developer "shall review and, as appropriate, update its frontier AI framework at least once per year" (California SB 53, 22757.12(b)(1))
Counter-examples
Not to be confused with
- A system card or model card is a per-model document produced under a safety framework, not the framework itself.
- A general responsible-AI principles page with no thresholds, no severe-risk scope and no review cadence is not a safety framework.
Related terms
Related AI safety terms
Terms defined under this element that are not NIKOLAI elements themselves, each as the cited document uses it. Every term is also listed in the AI safety glossary.
- Responsible Scaling Policy (RSP)
- Responsible Scaling Policy (RSP) is Anthropic's safety framework, its voluntary framework for managing catastrophic risks from advanced AI systems. Version 3.4, effective July 8, 2026, maps capability thresholds to mitigations and requires periodic Risk Reports.
- Source: Anthropic Responsible Scaling Policy v3.4, effective 8 Jul 2026
- Preparedness Framework
- Preparedness Framework is OpenAI's safety framework: its approach to tracking and preparing for frontier capabilities that create new risks of severe harm. Version 2 (April 2025) sets High and Critical thresholds for three Tracked Categories.
- Source: OpenAI Preparedness Framework v2
- Frontier Safety Framework
- Frontier Safety Framework is Google DeepMind's safety framework, covering severe risks that may arise from the high-impact capabilities of frontier AI models. Version 3.1 (April 2026) defines Critical Capability Levels across four risk domains.
- Source: Google DeepMind Frontier Safety Framework v3.1
- Frontier AI framework
- Frontier AI framework is the California SB 53 term for documented technical and organizational protocols to manage, assess and mitigate catastrophic risks. Large frontier developers must publish one, comply with it and review it at least yearly.
- Source: California SB 53
- Advanced AI Scaling Framework
- Advanced AI Scaling Framework is Meta's safety framework. Version 2, dated April 7, 2026, succeeded the Frontier AI Framework of February 2025 and covers chemical and biological, cybersecurity and loss-of-control risks.
- Source: Meta Advanced AI Scaling Framework v2
- Frontier safety policy
- Frontier safety policy is METR's generic name for protocols adopted by leading AI companies to keep the risks of developing and deploying state-of-the-art AI models at an acceptable level; METR's December 2025 edition listed twelve companies with one.
- Source: METR, Common Elements of Frontier AI Safety Policies
- Safety and Security Framework
- Safety and Security Framework is the EU General-Purpose AI Code of Practice term for the document outlining the systemic risk management processes and measures a signatory uses to keep risks from its models acceptable.
- Source: EU GPAI Code of Practice, Safety and Security Chapter
Source of record
Where this definition comes from
California SB 53, 22757.11(g); 22757.12(b)(1)
“"Frontier AI framework" means "documented technical and organizational protocols to manage, assess, and mitigate catastrophic risks." A large frontier developer "shall review and, as appropriate, update its frontier AI framework at least once per year."”
https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB53EU GPAI Code of Practice, Safety and Security Chapter, Commitment 1
“"Signatories commit to adopting a state-of-the-art Safety and Security Framework (“Framework”). The purpose of the Framework is to outline the systemic risk management processes and measures that Signatories implement to ensure the systemic risks stemming from their models are acceptable."”
https://ec.europa.eu/newsroom/dae/redirection/document/118119Frontier AI Safety Commitments, AI Seoul Summit 2024, Preamble, paragraph 1
“Signatories undertake "to demonstrate how they have achieved this by publishing a safety framework focused on severe risks by the upcoming AI Summit in France."”
https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024
Crosswalk
How named organisations use this concept
| Organisation | Their term, as published | Match & verification | Source |
|---|---|---|---|
| AnthropicShadow mapping Responsible Scaling Policy v3.4 | “"Our Responsible Scaling Policy (RSP) is our voluntary framework for managing catastrophic risks from advanced AI systems." (Introduction, p.3). "We have always intended for our RSP to be a living document." (p.3)” Anthropic uses "catastrophic risk" in its plain meaning and addresses statutory definitions such as SB 53's "in separate compliance frameworks" (p.4, fn.1). | exactEQ confidence: high | Anthropic Responsible Scaling Policy v3.4, effective 8 Jul 2026 |
| OpenAIShadow mapping Preparedness Framework v2 | “OpenAI's framework is its "approach to tracking and preparing for frontier capabilities that create new risks of severe harm" (p.1). "Version 2. Last updated: 15th April, 2025" (cover).” OpenAI also publishes a separate Frontier Governance Framework, which serves as its TFAIA frontier AI framework and as the public summary of its EU Safety and Security Framework (FGF §1). | exactEQ confidence: high | OpenAI Preparedness Framework v2 |
| Google DeepMindShadow mapping Frontier Safety Framework v3.1 | “Protocols for "severe risks that may arise from the high-impact capabilities of frontier AI models" (p.2). "The Frontier Safety Framework will be reviewed at least once a year" (s.5.1, p.17).” | exactEQ confidence: high | Google DeepMind Frontier Safety Framework v3.1 |
| MetaShadow mapping Advanced AI Scaling Framework v2 | “Covers catastrophic risks in three areas, "Chemical & Biological, Cybersecurity, and Loss of Control" (preamble, p.2); Appendix II change log records v2.0 (April 7, 2026) and the initial version, titled Frontier AI Framework (February 3, 2025) (p.44).” | exactEQ confidence: high | Meta Advanced AI Scaling Framework v2 |
| xAIShadow mapping Frontier AI Framework, 30 Jun 2026 | “"This Frontier AI Framework ("FAIF") outlines xAI's approach to policies for mitigation of significant risks associated with the development, deployment, and release of xAI's frontier AI models, such as Grok." (s.1). "Effective Date 30 June 2026".” This source's PDF metadata /Title reads "Privileged/Confidential DRAFT working FRAMEWORK DOC"; no xAI statement disambiguating draft vs. final status was found, so its status remains unresolved. | exactEQ confidence: medium | xAI Frontier AI Framework, 30 Jun 2026 |
| AmazonShadow mapping Amazon's Frontier Model Safety Framework | “"We will therefore revisit this Framework at least annually and update it as necessary" (s.4, p.5).” The PDF carries no date; CASRAI has not verified the publication date or any later revision. | exactEQ confidence: medium | Amazon's Frontier Model Safety Framework |
| EUShadow mapping Safety and Security chapter | “"Signatories commit to adopting a state-of-the-art Safety and Security Framework (“Framework”). The purpose of the Framework is to outline the systemic risk management processes and measures that Signatories implement to ensure the systemic risks stemming from their models are acceptable." (Commitment 1)” | exactEQ confidence: high | EU GPAI Code of Practice, Safety and Security Chapter |
| California SB 53Shadow mapping SB 53 statute text | “"Frontier AI framework": "documented technical and organizational protocols to manage, assess, and mitigate catastrophic risks" (22757.11(g)). A large frontier developer "shall write, implement, comply with, and clearly and conspicuously publish on its internet website a frontier AI framework" (22757.12(a)).” | exactEQ confidence: high | California SB 53 |
| AI Seoul Summit (Frontier AI Safety Commitments)Shadow mapping Frontier AI Safety Commitments | “Signatories undertake to demonstrate how they meet the commitments "by publishing a safety framework focused on severe risks by the upcoming AI Summit in France" (P1); the framework is "their safety framework (as described under I-VI)" (VIII).” A voluntary commitment to publish a framework, not a definition of one; the content is described only by reference to Commitments I-VI. | closeCL confidence: high | Frontier AI Safety Commitments, AI Seoul Summit 2024 |
| METRShadow mapping Common Elements of Frontier AI Safety Policies | “"Frontier safety policies": "protocols adopted by leading AI companies to ensure that the risks associated with developing and deploying state-of-the-art AI models are kept at an acceptable level." (/common-elements, Introduction)” METR states the analysis "is descriptive, not prescriptive, and does not represent METR's recommendations" (footnote 1). | exactEQ confidence: high | METR, Common Elements of Frontier AI Safety Policies |
What do these codes mean?
- exact
- The source term is equivalent to this element
- close
- The source term is close but not equivalent to this element
- broad
- The source term is broader than this element
- narrow
- The source term is narrower than this element
- none
- No mapping claim — used for false-friend and declared-but-undefined rows
- EQ
- Equivalent
- CL
- Close
- BR
- Source is broader than the element
- NR
- Source is narrower than the element
- FF
- False friend — same or similar label, different meaning
- DU
- Declared but undefined by the source
- UV
- Unverified
Related, not mapped
Pointers that are not crosswalk claims
These sources mention this concept but do not define or map it clearly enough to count as a crosswalk row — noted here so the research is visible without overstating it as a mapping.
- International AI Safety Report 2026 (Chair: Yoshua Bengio)
"In 2025, 12 companies published or updated their Frontier AI Safety Frameworks", described as "documents that describe how they plan to manage risks as they build more capable models"; "AI risk management initiatives remain largely voluntary" (official publication page, fetched 2026-11-13).
International AI Safety Report 2026
Divergence
Where sources materially disagree
The same kind of document goes by many names: Responsible Scaling Policy (Anthropic), Preparedness Framework (OpenAI), Frontier Safety Framework (Google DeepMind), Advanced AI Scaling Framework (Meta, formerly Frontier AI Framework), Frontier AI Framework (xAI), Frontier Model Safety Framework (Amazon), frontier AI framework (California SB 53), Safety and Security Framework (EU Code) and frontier safety policy (METR). Scope differs too: SB 53 frames the document around "catastrophic risks", the EU Code around "systemic risks", the Seoul Commitments around "severe risks". OpenAI keeps two documents, a Preparedness Framework and a Frontier Governance Framework, the second serving its statutory and EU obligations; Anthropic likewise handles statutory definitions "in separate compliance frameworks".
Gap
Not every framework carries its own version and date: OpenAI's Frontier Governance Framework has no date or version number in the PDF, and xAI's June 2026 framework has an effective date but no version number and draft-marked PDF metadata. A framework instance cannot be cited precisely without both.
