Laws compared
A cell records what a text requires of the parties it defines. It does not show that any developer falls within them, or complied.
What does each binding government instrument that addresses developers of frontier or general-purpose AI models say on the same questions? Each column is one instrument; each row is one question, worded so that it fits every instrument. The questions are Selekoa’s; an instrument’s own terms appear only in quotation marks. Some instruments address providers of services rather than developers, and the first row quotes the party each one defines.
Question Q06Outside assessment
Does the instrument require, or provide for, assessment of the model or of the developer’s practices by anyone other than the developer?
Criterion: Stated if the instrument requires an assessment by an independent or external party or by a public authority. Partly if it only permits or empowers one (an authority “may” evaluate or request access), or provides for an assessment of filings or documents rather than of the model.
What the statuses mean
- Stated
- A passage of the instrument meets the question’s criterion as worded, however conditional or narrow its scope. The condition or scope is in the note.
- Stated, partly
- A passage has what the criterion asks for, but in a weaker form than the criterion’s own words: for example, a power (“may”) where the criterion asks for a duty (“shall”), or a duty without the deadline the question asks about.
- Not found in the recorded search
- No passage meeting the criterion was found. The cell lists the sections read, the search terms and every passage set aside, with the reason. It does not show that the instrument lacks such a provision elsewhere, or that no other law imposes one.
- Does not apply
- The question does not arise for this instrument, for the reason given (for example, there is no framework document whose publication could be asked about).
Four rules decide “partly” in borderline cases:
- An order to an agency to impose a duty counts as partly.
- Protection only by reference to another law counts as partly.
- “Without undue delay” counts as a stated deadline.
- Required security measures for model weights count as risk-management measures under Q03.
Every stated or partly cell quotes the instrument’s own text with its article, section or page; Chinese passages carry Selekoa’s working translation. Three cells give a paraphrase with its locator instead, and say so. A not-found cell lists the sections read, the terms searched and every passage set aside, with the reason.
Paraphrased with its locator, not quoted: quoting it would pass the source’s quotation limit, or the source’s length is not known.
Limits of this comparison
- A cell records what the text requires of the parties it defines. It does not show that any named developer falls within them, or that anyone complied.
- A not-found cell is bounded by its search: the sections read and the terms searched. Another provision, or another law, may still impose the duty.
- The Code of Practice is a compliance route, not a law. Executive Order 14110 is revoked, and the BIS rule’s rescission was announced; each column shows its standing.
- The table compares and does not rank. Statuses are not scores and are not to be summed.
- Biological and chemical risk provisions are recorded at policy level only.
- Dates keep their kind: adoption, signature, approval, publication, entry into force and application are different events. A computed date is marked as computed.
How we searched
The instruments started from those on the frameworks-and-law page. A column was kept where the instrument is binding, or is the AI Act’s stated means of compliance (the Code), and places duties on developers or providers of frontier or general-purpose AI models. Fifty other instruments on that page were set aside, each with its reason, in the research pack: proposed or vetoed texts, guidance and reports, voluntary commitments, binding texts that place no duty on developers, and one standard whose text was not retrieved (GB 45438-2025).
United Kingdom: within the searches made, no UK law or other binding instrument placing duties on developers of such models was located. This is not a finding that none exists. The 2024 King’s Speech announced an intent to legislate; the 2026 briefing notes contain none of the searched terms; the memoranda with developers state that they are not legally binding; and title searches of legislation.gov.uk for 2023 to 2026 found one instrument, which places a duty on the Information Commissioner.
Other US federal texts: as read, none places a binding duty on developers. Executive Order 14409 of 2 June 2026 provides for a voluntary access framework and says that its section 3 authorises no licensing requirement.
Not yet read: the Republic of Korea’s AI Basic Act, Japan’s AI Promotion Act (2025), China’s provisions on deep synthesis (2022) and on algorithm recommendation (2021), and Colorado SB 24-205 (2024). They are not in the dataset; a later pack starts with Korea.
Illinois: the act’s official website refused the preparer’s requests, and the Internet Archive reset the connections. The cross-family checker’s session saved the enacted text from an Internet Archive capture of the official PDF (2026-09-08), and the cells were drafted from that saved text. The two model families read the act independently and agreed on the status of 13 of 15 cells. The approval date rests on the checker’s copy of the official bill status record, which the preparer could not open; it is given as a paraphrase, labelled as that copy.
Texts were read in official copies: the EU texts through the Publications Office’s repository, because EUR-Lex returned empty responses; California SB 53 from the Legislative Counsel’s database file, because its website refused requests; New York’s chapters in the Senate print and the Assembly’s bill records, not the session laws; the Chinese measures in the gov.cn policy library. Six web pages served changed bytes between retrievals; every quotation still matched.
Roles: one model family (Claude) drafted the cells, audited 50 of them in a fresh context and adjudicated. A model from another family (GPT, through Codex) checked 122 items, answering 85 of them blind before seeing the drafted answers. Where it disagreed, its status was applied. One cell it disputed (Executive Order 14110, Q06) was settled as not found after its second pass. For Illinois the roles were reversed: the checker read the act first, blind, and the drafter then wrote all fifteen cells without seeing its answers. The maintainer decided the remaining questions, including the four rules for “partly” and Illinois Q05 (partly, as for California’s identical wording). No person has checked the cells against the texts.
European Union
EU AI Act
Stated, partly
“The AI Office, after consulting the Board, may conduct evaluations of the general-purpose AI model concerned”
Article 92(1), chapeau
“through APIs or further appropriate technical means and tools, including source code”
Article 92(3)
Note: A power, not a requirement: the AI Office may evaluate a model to assess compliance where information under Article 91 is insufficient, or to investigate systemic risks; under Article 92(2) the Commission may appoint independent experts to carry out evaluations. Article 55(1)(a) requires adversarial testing but does not require that it be done by an external party. Search for the rest of the criterion: Articles 55, 56, 88 to 94; Annex XI; Regulation (EU) 2026/1744 Article 1, point (21). Terms searched: “independent”, “external”, “third party”, “evaluat”, “audit”. Passages read and not counted: Annex XI, Section 2, point 2: Asks for a description of internal and/or external adversarial testing 'where applicable'; does not require external testing. Article 56(6) as replaced by Regulation (EU) 2026/1744, Article 1, point (21): Commission assessment of the adequacy of codes of practice, not of a model or a provider's practices.
Inspect sources and checks: EU AI Act, Q06 Outside assessmentEU Code of Practice, Safety and Security
Stated
“Signatories will ensure that adequately qualified independent external evaluators conduct model evaluations”
Appendix 3.5, first paragraph (PDF p. 40)
Note: Subject to two exceptions in Appendix 3.5: the model is a 'similarly safe or safer model', or the Signatory fails to appoint qualified evaluators despite early search efforts. Measure 3.5 separately provides access for independent external evaluators for post-market monitoring, and Appendix 4.5 refers to security reviews by an independent external party.
Inspect sources and checks: EU Code of Practice, Safety and Security, Q06 Outside assessment
United States: states
California SB 53
Stated, partly
Paraphrase: The Office of Emergency Services must review the critical safety incident reports that frontier developers submit.
SEC. 2, Bus. & Prof. Code § 22757.13(d)
Note: The Office of Emergency Services shall review critical safety incident reports submitted by frontier developers: a review of filings, not of the model. The framework must describe how the developer approaches using third parties to assess catastrophic risks (§ 22757.12(a)(5)), and the transparency report summarises the extent to which third-party evaluators were involved (§ 22757.12(c)(2)(C)); neither requires an outside assessment. Search for the rest of the criterion: SEC. 2 in full; SEC. 3 (Gov. Code § 11546.8); SEC. 4. Terms searched: “third part”, “independent”, “audit”, “review”, “evaluat”. Passages read and not counted: SEC. 2, Bus. & Prof. Code § 22757.12(a)(5): Describes the content of the developer’s framework (its approach to using third parties); does not require an outside assessment. SEC. 2, Bus. & Prof. Code § 22757.12(c)(2)(C): Disclosure of the extent of third-party evaluator involvement; involvement is not required. SEC. 2, Bus. & Prof. Code § 22757.14(b)(5): Concerns external verifiability of coverage definitions in the Department of Technology’s recommendations, not assessment of a model.
Inspect sources and checks: California SB 53, Q06 Outside assessmentNew York RAISE Act (As amended, Chapter 96 of 2026)
Stated, partly
“The office shall review critical safety incident reports submitted”
Bill text, § 2 (GBL § 1422(4) as re-added), p. 6, lines 24–25
Note: The office (within the department of financial services, § 1420(16)) reviews incident reports: a review of filings, not of the model. The framework describes the developer’s use of third parties to assess catastrophic risks (§ 1421(1)(e)) and the transparency report the extent of third-party evaluator involvement (§ 1421(3)(b)(iii)); neither requires an outside assessment. Search for the rest of the criterion: Bill text §§ 1–5. Terms searched: “third part”, “independent”, “audit”, “review”, “evaluat”. Passages read and not counted: GBL § 1421(1)(e) as re-added, p. 4: Content of the developer’s framework, not a required outside assessment. GBL § 1428(2) as re-added, p. 8: Disclosure statement on ownership and contacts, filed with the office; not an assessment of the model or of risk practices.
Inspect sources and checks: New York RAISE Act (As amended, Chapter 96 of 2026), Q06 Outside assessmentNew York RAISE Act (As signed, Chapter 699 of 2025)
Stated, partly
“Grant the attorney general and division of homeland security and emergency services or the attorney general access to the safety and security protocol”
Proposed GBL § 1421(1)(c)(ii), p. 4, lines 22–25
Note: Access is granted “upon request”, with redactions only as required by federal law: access to a document, with no assessment by the authority required. § 1421(1)(d) requires test records detailed enough for third parties to replicate the testing procedure but does not require a third-party assessment. Search for the rest of the criterion: Whole print S6953-B, pp. 1–5. Terms searched: “third part”, “independent”, “audit”, “access”, “request”. Passages read and not counted: Proposed GBL § 1421(1)(d), p. 4: Replicability of tests by third parties is a record-keeping standard; no outside assessment is provided for. Proposed GBL § 1420(12)(d), p. 3: Protocol must enable the developer “or third party” to comply; not an outside assessment.
Inspect sources and checks: New York RAISE Act (As signed, Chapter 699 of 2025), Q06 Outside assessmentIllinois AI Safety Measures Act
Stated
“a large frontier developer shall annually retain a third party to perform an independent audit of compliance with the requirements of this Section.”
Section 10(d), first sentence, PDF p. 8
Note: An annual independent audit of the large frontier developer's compliance with Section 10 (framework, transparency report, internal-use summaries, statements, redactions): an assessment of the developer's practices, not a test of the model. It begins on 1 January 2028 or 90 days after the developer first qualifies as a large frontier developer, whichever is later. The auditor must have no financial interest in the developer (or the developer in it), must have access to unredacted materials, reports on substantial compliance and internal controls (Section 10(d)(1)–(2)); a redacted copy and a summary are published and sent to the Agency and the Attorney General within 30 days (Section 10(d)(4)). The Agency and the Attorney General also review incident reports (Section 15(d)), a review of filings.
Inspect sources and checks: Illinois AI Safety Measures Act, Q06 Outside assessment
China
These measures address providers of services, not developers as such.
China, generative AI services (2023)
Stated, partly
“开展监督检查”
Working translation: carry out supervision and inspection [of generative AI services]
Article 19 (第十九条), first paragraph
“按要求对训练数据来源、规模、类型、标注规则、算法机制机理等予以说明”
Working translation: explain, as required, the sources, scale and types of training data, annotation rules, algorithmic mechanisms and principles, and so on
Article 19 (第十九条), first paragraph
Note: Competent authorities (有关主管部门) carry out supervision and inspection of generative AI services 'according to their duties'; providers shall cooperate, explain the listed matters as required and provide necessary technical and data support. The provision is an inspection of the service with explanations from the provider; it does not require an assessment of the model by an outside party. Article 19(2) binds bodies and personnel taking part in security assessment and inspection to confidentiality, without stating who conducts the Article 17 assessment. Search for the rest of the criterion: Articles 16–21. Terms searched: “监督检查”, “评估”, “第三方|机构”. Passages read and not counted: Article 17 (第十七条): Security assessment is a duty of the provider 'in accordance with relevant state provisions'; the Measures do not say an outside party conducts it.
Inspect sources and checks: China, generative AI services (2023), Q06 Outside assessmentChina, AI content labelling (2025)
Stated, partly
“互联网应用程序分发平台应当核验其生成合成内容标识相关材料”
Working translation: the internet application distribution platform shall verify its materials relating to the labelling of generated synthetic content
Article 7 (第七条)
“服务提供者在履行算法备案、安全评估等手续时，应当按照本办法提供生成合成内容标识相关材料”
Working translation: when completing procedures such as algorithm filing and security assessment, service providers shall provide materials relating to the labelling of generated synthetic content in accordance with these Measures
Article 12 (第十二条)
Note: App distribution platforms verify the labelling materials of app service providers that offer AI generation or synthesis services, at listing or launch review (Article 7); labelling materials are supplied in algorithm filing and security assessment procedures under other rules (Article 12). Both concern labelling documents, not an assessment of the model or of risk practices. Search for the rest of the criterion: Articles 6–13. Terms searched: “核验”, “评估”, “备案”.
Inspect sources and checks: China, AI content labelling (2025), Q06 Outside assessment
Revoked, or rescission announced
Kept as part of the record, each with its standing shown.
US Executive Order 14110
Not found in the recorded search
- Read
- Whole cached text of the Federal Register printing (88 FR 75191–75226), with close reading of Secs. 3, 4.1, 4.2, 12 and 13 and the signature block.
- Terms searched
- “evaluat”, “assess”, “audit”, “third-party”, “independent”, “model capabilities”.
- Passages read and set aside
- Sec. 4.1(b), 88 FR 75196–75197: Directs the Secretary of Energy to develop AI model evaluation tools and testbeds; not an assessment of a covered company's model. Sec. 4.1(a)(i)(C), 88 FR 75196: Initiative to create guidance and benchmarks for evaluating and auditing AI capabilities; guidance, not an assessment. Sec. 4.4(a)(i), including (A), 88 FR 75200: Directs the Secretary of Homeland Security to evaluate the potential for AI to be misused for CBRN threats and, under (A), to consult experts including third-party model evaluators “to evaluate AI model capabilities to present CBRN threats”: a policy-level evaluation by a public authority, not an assessment of a covered company’s model or practices. Read and set aside by maintainer decision 6a: the cross-family checker, in its second pass of 2026-11-03, agreed that this passage does not meet Q06.
Note: No provision requiring or providing for assessment of a covered company's model or practices by an authority or third party was found. Agency evaluation tools and guidance are set aside below.
Open the search record: US Executive Order 14110, Q06 Outside assessmentUS BIS AI diffusion rule
Stated, partly
“attested annually by a Third-Party Assessment Organization (3PAO) and made available to BIS”
Supplement No. 10 to part 748, para. 14(b), 90 FR 4566 (PDF p. 23)
“cooperate with BIS’s auditing of records and facilities”
Supplement No. 10 to part 748, para. 11, 90 FR 4566 (PDF p. 23)
Note: Required annual third-party attestation of the data-centre NIST 800–53 compliance, and BIS auditing of VEU records and facilities. These assess the facility's security practices, not the model or a developer's practices as such. Partly because the assessed party is a data-centre end user under a VEU authorisation, not necessarily a model developer, and the subject is facility security.
Inspect sources and checks: US BIS AI diffusion rule, Q06 Outside assessment