Examples
Examples from published documents
- "On approximately an annual basis, we will commission a third-party review that assesses whether we adhered to this policy's main procedural commitments." (Anthropic RSP v3.4, section 4(7))
- "G42 will engage in annual external audits to verify compliance with the Framework." (G42 Frontier Safety Framework, section 5)
- "OpenAI will complete a Framework Assessment at least once every 12 months from the Effective Dates of the TFAIA and the EU Code of Practice" (OpenAI Frontier Governance Framework, section 7.2)
Counter-examples
Not to be confused with
- An external review of a risk report judges whether the risk assessment is sound; an adherence review asks whether the developer followed its own procedures. Anthropic's annual third-party review "will focus on procedural compliance, not substantive outcomes" (Anthropic RSP v3.4, section 4(7)).
- Updating a framework is not the same as checking adherence to it, although the EU Code of Practice and Google DeepMind run both inside one assessment.
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.
- Framework assessment
- A Framework assessment, under the EU GPAI Code of Practice, is a review held at least every 12 months, or sooner on reasonable grounds, of whether a provider's Safety and Security Framework is adequate and whether the provider has adhered to it.
- Source: EU GPAI Code of Practice, Safety and Security Chapter
- Internal audit
- Internal audit is a company's own assurance function that checks whether its risk management processes work as described. The EU GPAI Code of Practice names an internal audit function as an example of an internal assurance provider for systemic risk.
- Source: EU GPAI Code of Practice, Safety and Security Chapter
- Third-party compliance audit
- A third-party compliance audit is an independent auditor's examination of whether a developer has substantially complied with its own frontier AI framework. The proposed US FRONTIER Act (H.R. 9925) would require one each year, signed by a lead auditor.
- Source: FRONTIER Act, 119th Congress (H.R. 9925, introduced, not enacted)
- Attestation
- An attestation is a formal statement by a named party that a specific fact holds, such as compliance with a framework. Anthropic's legislative proposal would require covered developers to certify annual compliance with their safety framework to a government agency.
- Source: Anthropic Advanced AI Framework (legislative proposal)
- Deviation
- A deviation, or non-adherence, is an instance in which a developer did not follow its own safety framework. The EU GPAI Code of Practice asks each Framework assessment to record any instances of, and reasons for, non-adherence and to make remediation plans.
- Source: EU GPAI Code of Practice, Safety and Security Chapter
Source of record
Where this definition comes from
Anthropic Responsible Scaling Policy v3.4, §4(7), p.16
“Procedural compliance review: On approximately an annual basis, we will commission a third-party review that assesses whether we adhered to this policy's main procedural commitments. This review will focus on procedural compliance, not substantive outcomes.”
https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdfEU GPAI Code of Practice, Safety and Security Chapter, Commitment 1, Measure 1.3
“Framework adherence: an assessment focused on the Signatories' adherence to the Framework, including: (a) any instances of, and reasons for, non-adherence to the Framework since the last Framework assessment”
https://ec.europa.eu/newsroom/dae/redirection/document/118119FRONTIER Act, 119th Congress (H.R. 9925, introduced, not enacted), Sec. 4(c)
“a large frontier developer shall retain a third party to perform an independent audit of compliance with the frontier AI framework of such developer”
https://www.govinfo.gov/content/pkg/BILLS-119hr9925ih/html/BILLS-119hr9925ih.htm
Crosswalk
How named organisations use this concept
Every row below is a shadow mapping. A shadow row is CASRAI's own reading of a published document. No lab, evaluator or regulator named on a shadow row has declared, endorsed, or been consulted on it. That changes only when an organisation files its own Mapping Declaration.
| Organisation | Their term, as published | Match & verification | Source |
|---|---|---|---|
| AnthropicShadow mapping Anthropic RSP v3.4 | “"Internal review: We will regularly conduct an internal review of our implementation of this policy." (§4(6)). "Procedural compliance review: On approximately an annual basis, we will commission a third-party review that assesses whether we adhered to this policy's main procedural commitments. This review will focus on procedural compliance, not substantive outcomes." (§4(7))” | exactEQ confidence: high | Anthropic Responsible Scaling Policy v3.4 |
| EUShadow mapping EU GPAI Code of Practice, Safety and Security Chapter | “Measure 1.3: a Framework assessment covers "Framework adherence: an assessment focused on the Signatories' adherence to the Framework, including: (a) any instances of, and reasons for, non-adherence to the Framework since the last Framework assessment"; "If point(s) (a) and/or (b) give rise to risks of future non-adherence, Signatories will make remediation plans as part of their Framework assessment." Measure 8.1 assurance levels: "(4) if available, internal assurance providers (e.g. an internal audit function); and (5) if available, external assurance providers (e.g. third-party auditors)."” | exactEQ confidence: high | EU GPAI Code of Practice, Safety and Security Chapter |
| OpenAIShadow mapping OpenAI Frontier Governance Framework | “"OpenAI will complete a Framework Assessment at least once every 12 months from the Effective Dates of the TFAIA and the EU Code of Practice" (§7.2). "For the EU Code of Practice, if we identify any instances of non-adherence or any measures that are required to be implemented to ensure continued adherence, we will draft and implement a remediation plan, and update the FGF as appropriate." (§7.2)” The non-adherence and remediation language is scoped to the EU Code of Practice; the assessment is run by OpenAI itself. | exactEQ confidence: high | OpenAI Frontier Governance Framework |
| Google DeepMindShadow mapping Frontier Safety Framework v3.1 | “The annual review "will involve (i) an assessment of the Framework's appropriateness for the management of significant and severe risk ... and (ii) an assessment of our adherence to the Framework." "The updated version and framework assessment will be reviewed by the appropriate corporate governance bodies." (s.5.1)” Internal assessment; no third-party reviewer is named. | exactEQ confidence: high | Google DeepMind Frontier Safety Framework v3.1 |
| MetaShadow mapping Meta Advanced AI Scaling Framework v2 | “"Meta's internal governance function periodically reviews our risk management practices and provides compliance oversight for product teams." (§2.3). "Framework updates will cover both the Framework's adequacy ... and adherence to it." (§5.1)” CL: internal compliance oversight and adherence covered within framework updates; no separate adherence report is described. | closeCL confidence: medium | Meta Advanced AI Scaling Framework v2 |
| MicrosoftShadow mapping Microsoft Frontier Governance Framework (Feb 2026) | “"This framework is subject to Microsoft's broader corporate governance procedures, including independent internal audit and board oversight." (s.5, p.9)” BR: internal audit applies through general corporate governance; scope and cadence for this framework are not stated. | broadBR confidence: medium | Microsoft Frontier Governance Framework (Feb 2026) |
| G42Shadow mapping G42 Frontier Safety Framework | “"Annual Governance Audits: G42 will have independent internal audits to verify compliance with our policy." "External Audits: To reinforce accountability, G42 will engage in annual external audits to verify compliance with the Framework." (s.5)” Auditor, standard and publication of results are not specified. | exactEQ confidence: medium | G42 Frontier Safety Framework |
| xAIShadow mapping xAI Frontier AI Framework (31 Dec 2025) | “Listed under transparency: "Frontier AI Framework Adherence: regularly review our adherence with this FAIF." (p.8). Governance: "Risk owners are also responsible for periodic audits to enforce framework implementation." (s.4)” CL: the review is stated within a list of items xAI may publish, with redaction allowed; reviewer and cadence are not defined. The research did not find this item in the June 2026 version. | closeCL confidence: medium | xAI Frontier AI Framework (31 Dec 2025) |
| US Congress (H.R. 9925, FRONTIER Act, not enacted)Shadow mapping FRONTIER Act (H.R. 9925) | “Sec. 4(c) Compliance Audit: annually "a large frontier developer shall retain a third party to perform an independent audit of compliance with the frontier AI framework of such developer"; the report includes "A description of whether the large frontier developer has substantially complied with the frontier AI framework of such developer", "a description of material deviations from such framework, an explanation of any deviation and its rationale", and "The signature of the lead auditor certifying the results of the audit."” Introduced bill, not enacted. A summary and redacted report would be published within 30 days of receipt. | exactEQ confidence: medium | FRONTIER Act, 119th Congress (H.R. 9925, introduced, not enacted) |
| AnthropicShadow mapping Anthropic Advanced AI Framework (legislative proposal) | “"Annual certification. Require Covered Developers to certify annual compliance with their safety framework to the Agency, with civil penalties for material misrepresentation." (p.5)” CL: a proposed self-certification to a regulator rather than a review; a policy proposal for government, not a company practice. | closeCL confidence: medium | Anthropic Advanced AI Framework (legislative proposal) |
| Seoul Summit signatoriesShadow mapping Frontier AI Safety Commitments (Seoul 2024) | “Commitment VIII: "Explain how, if at all, external actors, such as governments, civil society, academics, and the public are involved in the process of assessing the risks of their AI models and systems, the adequacy of their safety framework (as described under I-VI), and their adherence to that framework."” CL: asks signatories to explain any external role in assessing adherence; does not require a review. | closeCL 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
No audit duty, but non-adherence is penalised: a large frontier developer that "fails to comply with its own frontier AI framework shall be subject to a civil penalty" of up to $1,000,000 per violation (22757.15(a)), and may not make "a materially false or misleading statement about its implementation of, or compliance with, its frontier AI framework" (22757.12(e)(1)(B)).
California SB 53
Divergence
Where sources materially disagree
Sources split on who checks adherence: the developer itself (Google DeepMind, Meta, OpenAI's FGF assessment), internal audit (Microsoft, G42), a commissioned third party (Anthropic's annual procedural compliance review, G42's external audits, the proposed FRONTIER Act audit), or a regulator receiving a certification (Anthropic's legislative proposal). Scope also differs: Anthropic limits its third-party review to procedural compliance, while the EU Code and Google DeepMind pair adherence with an assessment of the framework's adequacy.
Gap
The research did not find a requirement for an independent adherence audit in California SB 53 or New York's RAISE Act; the proposed federal FRONTIER Act would add one. Few sources commit to publishing adherence-review results.
