Examples
Examples from published documents
- "In the extreme, organisations commit not to develop or deploy a model or system at all, if mitigations cannot be applied to keep risks below the thresholds." (Frontier AI Safety Commitments, Commitment IV)
- "We won't deploy these very capable models until we've built safeguards to sufficiently minimize the associated risks of severe harm." (OpenAI Preparedness Framework v2, p.1)
- "We will publish a preparedness report for each closed or open Frontier AI release" (Meta Advanced AI Scaling Framework v2, section 2.2.1)
Counter-examples
Not to be confused with
- A roadmap goal is not a commitment: Anthropic says its Frontier Safety Roadmap goals "are not hard commitments but rather public goals against which we will openly grade our progress" (Anthropic RSP v3.4, Introduction).
- A request addressed to someone else is not a commitment by its signers: the 'Pacing the Frontier' statement asks the U.S. government to act (Pacing the Frontier open letter).
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.
- Voluntary commitment
- A voluntary commitment is a pledge an AI developer makes without legal compulsion. Under the Frontier AI Safety Commitments, signatories undertook to develop and deploy frontier models responsibly and to publish a safety framework focused on severe risks.
- Source: Frontier AI Safety Commitments (AI Seoul Summit 2024)
- Statutory obligation
- A statutory obligation is a duty imposed by law rather than chosen by the developer. California SB 53, for example, requires a large frontier developer to write, implement, comply with and publish a frontier AI framework, with civil penalties for violations.
- Source: California SB 53
- Procedural versus outcome commitment
- A procedural commitment promises to follow a stated process, such as a review or approval step; an outcome commitment promises a result. Anthropic's annual third-party review checks "procedural compliance, not substantive outcomes."
- Source: Anthropic Responsible Scaling Policy v3.4
- Competitor-contingent commitment
- A competitor-contingent commitment is a pledge whose terms depend on what other developers do. Anthropic's RSP v3.4 sets different commitments for scenarios such as "Anthropic in the lead" and "Competitors have strong safety measures."
- Source: Anthropic Responsible Scaling Policy v3.4
Source of record
Where this definition comes from
Frontier AI Safety Commitments (AI Seoul Summit 2024), Preamble
“The above organisations, in furtherance of safe and trustworthy AI, undertake to develop and deploy their frontier AI models and systems responsibly, in accordance with the following voluntary commitments, and 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-2024OpenAI Preparedness Framework v2, p.1
“We won't deploy these very capable models until we've built safeguards to sufficiently minimize the associated risks of severe harm.”
https://cdn.openai.com/pdf/18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdfAnthropic Responsible Scaling Policy v3.4, Introduction, p.4
“We have also adopted a set of competitor-contingent commitments (see Appendix A) aimed at staying in line with these recommendations in scenarios where we can be confident that other relevant AI developers are doing the same.”
https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf
Crosswalk
How named organisations use this concept
| Organisation | Their term, as published | Match & verification | Source |
|---|---|---|---|
| AnthropicShadow mapping Anthropic RSP v3.4 / 'We Must Pace the Frontier' | “"Anthropic is unilaterally committing to this step now" (Embedded Evaluators); RSP v3.4 Introduction verbatim: "Our recommendations for industry-wide safety outline what it would take, at an industry-wide level, to keep catastrophic risks reliably low ... We also include our planned mitigations as a company," with "2. Frontier Safety Roadmap" and "3. Risk Reports" as separate top-level sections; the RSP is explicitly "our voluntary framework for managing catastrophic risks from advanced AI systems."” Also cites Amodei's 'We Must Pace the Frontier' essay for the unilateral-commitment framing; that essay and the 'Pacing the Frontier' open letter are two distinct documents. | exactEQ confidence: high | Anthropic Responsible Scaling Policy v3.4 |
| OpenAIShadow mapping OpenAI Preparedness Framework v2 | “"We won't deploy these very capable models until we've built safeguards to sufficiently minimize the associated risks of severe harm." (PF p.1).” | exactEQ confidence: high | OpenAI Preparedness Framework v2 |
| Google DeepMindShadow mapping Frontier Safety Framework v3.1 | “"aim to share relevant information with appropriate government authorities" (s.5.2): aspirational. The FSF is described as voluntary internal protocols.” | closeCL confidence: medium | Google DeepMind Frontier Safety Framework v3.1 |
| xAIShadow mapping xAI Frontier AI Framework (30 Jun 2026) | “Modality split: "xAI will conduct a full systemic risk assessment ... at least once a year" (firm) versus "may" for trigger-point evaluations, tiered availability, shutdown and post-mortems. Musk's posts carry no xAI commitment.” The xAI framework's PDF metadata /Title reads 'Privileged/Confidential DRAFT working FRAMEWORK DOC'; no xAI statement disambiguating draft vs. final was found. Also cites a post on X by xAI's founder, which carries no commitment. | closeCL confidence: medium | xAI Frontier AI Framework (30 Jun 2026) |
| MetaShadow mapping Meta Advanced AI Scaling Framework v2 | “"We will publish a preparedness report for each closed or open Frontier AI release" (§2.2.1); whistleblower protocol still "developing."” | closeCL confidence: medium | Meta Advanced AI Scaling Framework v2 |
| EUShadow mapping EU GPAI Code of Practice (landing page) | “Signatories adhere to all three chapters; "xAI signed up to the Safety and Security Chapter" only and must show other compliance "via alternative adequate means" (landing page, as reported).” | exactEQ confidence: high | EU GPAI Code of Practice (landing page) |
| US Government (NIST CAISI)Shadow mapping NIST/CAISI announcement | “CAISI "announced new agreements with Google DeepMind, Microsoft and xAI" (terms not public).” Terms of the agreements announced by CAISSI (formerly CAISI) are not public; CASRAI has not verified access level, days-before-release or publication rights against a primary text, and the full NIST announcement page returned a 404. Treat scope/modality as unconfirmed. | closeCL confidence: low | NIST/CAISI announcement |
| METRShadow mapping METR | “Common elements are phrased as "Intentions to ..." and "Commitments to ..." (element definitions).” | closeCL confidence: medium | METR |
| Seoul Summit signatoriesShadow mapping Frontier AI Safety Commitments (Seoul 2024) | “Signatories "undertake to develop and deploy their frontier AI models and systems responsibly, in accordance with the following voluntary commitments, and to demonstrate how they have achieved this by publishing a safety framework focused on severe risks by the upcoming AI Summit in France." Eight commitments (I-VIII) under three outcomes; Commitment IV: "In the extreme, organisations commit not to develop or deploy a model or system at all, if mitigations cannot be applied to keep risks below the thresholds."” Voluntary commitments by the listed organisations (UK and Republic of Korea governments' publication, updated 7 February 2025). The 'Pacing the Frontier' open letter formerly combined into this row is a request to the U.S. government, not a commitment by its signatories; it is now recorded as a related pointer. | exactEQ confidence: medium | Frontier AI Safety Commitments (AI Seoul Summit 2024) |
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.
- California SB 53
Statutory duties expressed with 'shall' rather than voluntary commitments (for example, 22757.12(a): a large frontier developer "shall write, implement, comply with, and clearly and conspicuously publish" a frontier AI framework). A related but non-identical concept: legal compulsion rather than a voluntary pledge; recorded as a pointer, not a mapping.
California SB 53 - Frontier Model Forum
One option floated, not adopted: to "commit to maintaining safety standards as stringent as the current ecosystem baseline, combined with sharing sufficient information, either publicly or privately, about implemented safeguards" (s3.2).
Frontier Model Forum, Risk Taxonomy and Thresholds - Pacing the Frontier (open letter)
"We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development." Signed as "a statement from 1,386 employees of frontier AI companies" in a personal capacity. A request to government rather than a commitment by the signers; useful as a precedent for identified-or-anonymous signatory records.
'Pacing the Frontier' open letter







