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 Q03Safety framework
Must the covered party write or adopt a document setting out how it assesses and manages risks from its models?
Criterion: A provision requiring the covered party to have a written framework, policy or protocol on the risks of its models. A policy on a single other subject (copyright, content labelling) does not meet it. Partly if a risk-management system or measures are required without a written document.
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
“assess and mitigate possible systemic risks at Union level”
Article 55(1), point (b)
Note: Providers of models with systemic risk must assess and mitigate systemic risks; Articles 53 and 55 do not require a written risk framework, policy or protocol. The only written policy required (Article 53(1)(c)) concerns copyright. Search for the rest of the criterion: Chapter V (Articles 51 to 56), Article 3 points (63) to (65), Annexes XI to XIII; Regulation (EU) 2026/1744 Article 1 (all points) for amendments to these provisions. Terms searched: “framework”, “policy”, “protocol”, “risk management”, “Article 5[1-6]”, “general-purpose AI model”. Passages read and not counted: Article 53(1), point (c): A policy on copyright and related rights, a single other subject. Article 9: Risk management system for high-risk AI systems, not general-purpose AI models. Annex XI, Section 2, point 1: Technical documentation content (description of evaluation strategies), not a risk framework.
Inspect sources and checks: EU AI Act, Q03 Safety frameworkEU Code of Practice, Safety and Security
Stated
“Signatories commit to adopting a state-of-the-art Safety and Security Framework (“Framework”).”
Commitment 1, first paragraph (PDF p. 6)
Note: Measure 1.1 lists the Framework's contents (trigger points, systemic risk acceptance criteria and tiers, responsibility allocation, update process); Measure 1.3 requires updates and a Framework assessment at least every 12 months. A commitment of Signatories.
Inspect sources and checks: EU Code of Practice, Safety and Security, Q03 Safety framework
United States: states
California SB 53
Stated
“A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its internet website a frontier AI framework”
SEC. 2, Bus. & Prof. Code § 22757.12(a), opening clause
“review and, as appropriate, update its frontier AI framework at least once per year.”
SEC. 2, Bus. & Prof. Code § 22757.12(b)(1)
Note: Applies to large frontier developers. “Frontier AI framework” is defined in § 22757.11(g) as documented technical and organizational protocols to manage, assess and mitigate catastrophic risks; § 22757.12(a)(1)–(10) lists the matters the framework describes, including thresholds, mitigations, third-party assessment, cybersecurity, incident response, internal governance and internal use.
Inspect sources and checks: California SB 53, Q03 Safety frameworkNew York RAISE Act (As amended, Chapter 96 of 2026)
Stated
“A large frontier developer shall … write, implement, comply with, and clearly and conspicuously publish on … its internet website a frontier AI framework”
Bill text, § 2 (GBL § 1421(1) as re-added), p. 4, lines 14–16
Note: The same wording is quoted in an earlier record of the dataset, split at the printed line numbers. “Frontier AI framework” is defined in § 1420(7) as documented technical and organizational protocols to manage, assess and mitigate catastrophic risks; § 1421(1)(a)–(j) lists what it describes “in detail”. Annual review under § 1421(2)(a).
Inspect sources and checks: New York RAISE Act (As amended, Chapter 96 of 2026), Q03 Safety frameworkNew York RAISE Act (As signed, Chapter 699 of 2025)
Stated
“Before deploying a frontier model, the large developer of such frontier model shall do all of the following: (a) Implement a written safety and security protocol”
Proposed GBL § 1421(1)(a), p. 4, lines 10–12
Note: “Safety and security protocol” is defined in § 1420(12) (p. 3) as documented technical and organizational protocols describing protections against critical harm, cybersecurity protections, a testing procedure, compliance and senior responsible personnel. § 1421(3) requires an annual review.
Inspect sources and checks: New York RAISE Act (As signed, Chapter 699 of 2025), Q03 Safety frameworkIllinois AI Safety Measures Act
Stated
“Beginning January 1, 2028, a large frontier developer … shall write, implement, comply with, and clearly and conspicuously publish on its website a frontier AI framework”
Section 10(a), opening clause, PDF pp. 4–5
“Frontier AI framework” means documented technical and organizational protocols to manage, assess, and mitigate catastrophic risks.”
Section 5, definition of “frontier AI framework”, PDF p. 3
Note: Applies to large frontier developers from 1 January 2028. Section 10(a)(1)–(10) lists the matters the framework describes, among them thresholds, mitigations, third-party assessment, cybersecurity for unreleased model weights, critical safety incidents and internal use; Section 10(b)(1) requires review and, as appropriate, update at least once a year. “…” in the quotation stands for the running footer at a PDF page break; no words of the act are omitted.
Inspect sources and checks: Illinois AI Safety Measures Act, Q03 Safety framework
China
These measures address providers of services, not developers as such.
China, generative AI services (2023)
Not found in the recorded search
- Read
- Order No. 15 preamble and Articles 1–24 in full (the page repeats the text; the first copy was read and the second spot-checked).
- Terms searched
- “标准”, “评估”, “制度|管理体系|风险”, “规则”, “措施”.
- Passages read and set aside
- Article 8 (第八条): Requires 'clear, specific and operable' data-annotation rules (标注规则) and annotation quality assessment: a document on a single other subject (data annotation), not on model risks. Article 4 (第四条), points (1)–(5): Requires 'effective measures' (有效措施) against prohibited content, discrimination and for transparency and accuracy, addressed to providing and using services; no risk-management system or written document on model risks is required, so not counted as partly. Point (2) requires effective measures against discrimination in algorithm design, training-data selection, model generation and optimisation and service provision: measures on a single other subject (discrimination), not a framework or system on model risks. Article 15 (第十五条): Complaint and reporting mechanism for the public; not a risk framework. Article 5 (第五条), second paragraph: Supports cooperation on risk prevention (风险防范) among organisations; encouragement, not a duty on providers.
Note: No provision requiring the provider to have a written framework, policy or protocol on the risks of its models was found.
Open the search record: China, generative AI services (2023), Q03 Safety frameworkChina, AI content labelling (2025)
Not found in the recorded search
- Read
- Notice 国信办通字〔2025〕2号 and Articles 1–14 in full.
- Terms searched
- “制度|管理体系|风险”, “评估”, “标准”.
- Passages read and set aside
- Article 8 (第八条): Service providers must state labelling methods and styles in the user service agreement (用户服务协议): a document on a single other subject (content labelling), excluded by the criterion. Article 9 (第九条): A user agreement (用户协议) setting the user's labelling obligations before unlabelled content is supplied: labelling only.
Note: No provision requiring a written framework, policy or protocol on model risks was found; the Measures concern content labelling only.
Open the search record: China, AI content labelling (2025), Q03 Safety framework
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
- “framework”, “policy”, “protocol”, “safety objectives”, “risk management”.
- Passages read and set aside
- Sec. 4.2(a)(i)(C), 88 FR 75197: Reporting of “a description of any associated measures the company has taken to meet safety objectives”; a report on measures, not a required written framework. Sec. 4.1(a)(i)(A), 88 FR 75196: Directs NIST to develop a companion resource to the AI Risk Management Framework; guidance, not a duty on companies (recorded under Q15). Sec. 4.4(b)(i), 88 FR 75201: OSTP framework for synthetic nucleic acid procurement screening; addressed to providers of nucleic acid sequences, not model developers.
Note: No provision directing that covered companies write or adopt a framework, policy or protocol on the risks of their models was found.
Open the search record: US Executive Order 14110, Q03 Safety frameworkUS BIS AI diffusion rule
Stated, partly
“Plans for the secure operation of data centers supporting a VEU shall be implemented in accordance with a security management defense-in-depth framework”
Supplement No. 10 to part 748, para. 14(e), 90 FR 4566 (PDF p. 23)
“The VEU will establish and bear responsibility for the following additional practices for AI security”
Supplement No. 10 to part 748, para. 15, 90 FR 4566 (PDF p. 23)
Note: Partly: the rule requires security measures, including AI-specific security practices covering model weights (para. 15), as conditions of validated end-user authorisation and, under § 740.27(a)(3)(ii), for facilities storing specified model weights. No written framework on the risks of models is required, and the data-centre plans are not a developer’s safety framework. The regulated party is the validated end user or facility, not necessarily a model developer. Status applied from the cross-family check. Required security measures for model weights count as risk-management measures (the fourth rule for “partly”, maintainer decision 5a of 2026-11-03). Search for the rest of the criterion: Federal Register PDF, 90 FR 4544–4584: DATES, SUPPLEMENTARY INFORMATION sections III and IV and Rulemaking Requirements; regulatory text §§ 740.27, 742.6(a)(13) and (b)(14), Supplement No. 10 to part 748 in full, ECCN 4E091 entry; signature and filing lines. Terms searched: “framework”, “plan”, “policy”, “safety”, “security requirements”. Passages read and not counted: § 740.27(a)(3)(ii), 90 FR 4559 (PDF p. 16): “stored in a facility that complies with paragraphs 14, 15 and 18”: extends the security conditions to facilities storing specified model weights; scope of the measures quoted above, not a written framework on model risks. Supplement No. 10 to part 748, para. 18(c), 90 FR 4567: Insider-threat programme; personnel security, recorded under this cell's note only by reference. Supplement No. 10 to part 748, para. 16, 90 FR 4567: Shipment security plan for chips in transit; not on models.
Inspect sources and checks: US BIS AI diffusion rule, Q03 Safety framework