selekoa

A documentary history of machine intelligence.

Records found and checked by machines, not reviewed by an editor · What this means →

Research question · 202 existing records

Frameworks, commitments and legal obligations, 2023–2026

From voluntary frameworks to legal obligations: how did frontier-AI developers’ self-governance change, 2023–2026?

Selekoa records on the developers’ own safety frameworks, the shared commitments of 2023 and 2024, the EU General-Purpose AI Code of Practice, the EU AI Act as published in 2024 and amended in 2026, California SB 53, the New York RAISE Act and one Chinese measure on generative AI services. It includes the Commission’s guidance on the AI Act and a draft rule on proceedings. Each record links to its claim, its evidence and the original source. The page lists records and adds no claim or conclusion of its own.

A machine selected and arranged these records. No human review of the selection or framing is recorded.

Download research starter

The chronology, every record in each part of this page, the open questions, the questions set aside and the list of what the records cannot answer. Each citation gives the original source and the place in it first, then the Selekoa claim and evidence identifiers. Stored paraphrases and permitted short quotes are included. Original documents are not.

Original sources for a reference manager. Every source this export cites, for Zotero, EndNote and similar tools. Cite the original documents, not Selekoa, and check each entry against them.

Related: Frontier safety frameworks · External commitments and obligations

Limits of this page
  • A framework, a commitment, a signature or a legal duty is never evidence of implementation, compliance or effectiveness. A document’s statement that it fulfils or complies with a text is that document’s own claim.
  • Records are placed side by side in date order. Order and closeness in date do not show that one led to another.
  • Signatures and framework links are listed only for the developers whose frameworks are recorded here. Google DeepMind publishes the frameworks, and the commitment lists name Google.
  • Framework revisions are limited to key revisions: each publisher’s new major version or new framework, or the later edition where editions carry no version number. The document histories page has every recorded edition and change.
  • Provisions are a limited, quoted selection under the quotation policy. They are not a summary of any text, and a text without a recorded provision has not been shown to lack it.
  • The EU AI Act records quote the Official Journal text of 2024, and the amendment records quote the text of Regulation (EU) 2026/1744. The consolidated Act was not inspected. The New York provisions are quoted from the legislature’s printed bill S8828 as shown on the Assembly’s site, and the 2025 chapter is known only through announcements. The Commission’s guidelines and template state its interpretation and are not binding. The Chinese measure is quoted in Chinese, and English statements about it are translations by a machine agent.
  • Open questions and questions set aside are reused from the commitments and document histories pages when they concern a record listed here.

Chronology

Issue, enactment and entry into force, and each framework edition that names a commitment or law. Each date keeps its basis. Date order is not a causal sequence.

  1. 2023-07-10 · publicationEnacted Interim Measures for the Management of Generative Artificial Intelligence Services (生成式人工智能服务管理暂行办法), Order No. 15Original source
  2. 2023-07-21 · publicationIssued Ensuring Safe, Secure, and Trustworthy AI: voluntary AI commitments (July 2023)Original source 1Archived copy, captured 2023-07-21Original source 2Archived copy, captured 2023-07-21
  3. 2023-08-15 · effectiveTakes effect Interim Measures for the Management of Generative Artificial Intelligence Services (生成式人工智能服务管理暂行办法), Order No. 15Original source
  4. 2023-09-12 · publicationIssued Voluntary AI Commitments (September 2023)Original source 1Archived copy, captured 2023-09-12Original source 2Archived copy, captured 2023-09-12
  5. 2023-09-19 · eventIssued Responsible Scaling Policy (2023)Original sourceArchived copy, captured 2024-01-18
  6. 2023-09-19 · effectiveNames a commitment or law RSP · 2023 → White House commitments (July 2023) (names)Original sourceArchived copy, captured 2023-09-19
  7. 2023-12-18 · publicationIssued OpenAI Preparedness Framework — BetaOriginal sourceArchived copy, captured 2023-12-18
  8. 2024-05-17 · eventIssued Frontier Safety Framework (2024)Original sourceArchived copy, captured 2025-11-04
  9. 2024-05-21 · publicationIssued Frontier AI Safety Commitments, AI Seoul Summit 2024Original source
  10. 2024-06-13 · signatureEnacted Regulation (EU) 2024/1689 (Artificial Intelligence Act), as published in the Official JournalOriginal sourceArchived copy, captured 2024-07-12
  11. 2024-10-15 · effectiveIssued Anthropic Responsible Scaling Policy — v2.0Original source 1Archived copy, captured 2024-10-22Original source 2Archived copy, captured 2026-09-18
  12. 2024-10-15 · effectiveNames a commitment or law RSP · v2.0 → Seoul commitments (2024) (states that it fulfils)Original sourceArchived copy, captured 2024-10-22
  13. 2024-10-15 · effectiveNames a commitment or law RSP · v2.0 → White House commitments (July 2023) (states that it fulfils)Original sourceArchived copy, captured 2024-10-22
  14. 2025-02 · publicationNames a commitment or law Microsoft FGF · Feb 2025 → Seoul commitments (2024) (states that it responds to)Original sourceArchived copy, captured 2025-02-26
  15. 2025-02 · publicationIssued Microsoft Frontier Governance Framework — Version 1 (February 2025)Original sourceArchived copy, captured 2025-02-26
  16. 2025-02-03 · publicationNames a commitment or law Meta FAF · v1.1 → Seoul commitments (2024) (states that it responds to)Original sourceArchived copy, captured 2025-02-13
  17. 2025-02-03 · publicationIssued Meta Frontier AI Framework — Version 1.1Original source 1Archived copy, captured 2025-02-03Original source 2Archived copy, captured 2026-04-08
  18. 2025-02-04 · publicationIssued Google DeepMind Frontier Safety Framework — v2.0Original sourceArchived copy, captured 2025-02-05
  19. 2025-03-31 · effectiveNames a commitment or law RSP · v2.1 → Seoul commitments (2024) (states that it fulfils)Original sourceArchived copy, captured 2025-04-02
  20. 2025-03-31 · effectiveNames a commitment or law RSP · v2.1 → White House commitments (July 2023) (states that it fulfils)Original sourceArchived copy, captured 2025-04-02
  21. 2025-04-15 · revisionIssued OpenAI Preparedness Framework — Version 2Original source 1Archived copy, captured 2025-04-15Original source 2Archived copy, captured 2025-04-16
  22. 2025-05-14 · effectiveNames a commitment or law RSP · v2.2 → Seoul commitments (2024) (states that it fulfils)Original sourceArchived copy, captured 2025-05-14
  23. 2025-05-14 · effectiveNames a commitment or law RSP · v2.2 → White House commitments (July 2023) (states that it fulfils)Original sourceArchived copy, captured 2025-05-14
  24. 2025-07-10 · publicationIssued General-Purpose AI Code of Practice (final version, 10 July 2025)Original source 1Archived copy, captured 2025-07-10Original source 2Archived copy, captured 2025-08-01
  25. 2025-07-18 · publicationIssued Communication to the Commission C(2025) 5045: approval of the content of the draft guidelines on the obligations of providers of general-purpose AI models (18 July 2025)Original source 1Original source 2
  26. 2025-09-22 · publicationIssued Google DeepMind Frontier Safety Framework — v3.0Original sourceArchived copy, captured 2025-09-22
  27. 2025-09-29 · signatureEnacted California SB 53 (2025): Transparency in Frontier Artificial Intelligence Act, as chapteredOriginal source 1Archived copy, captured 2025-09-30Original source 2Archived copy, captured 2025-12-11Original source 3Archived copy, captured 2025-09-29
  28. 2025-11-19 · publicationIssued Commission Guidelines on the scope of the obligations for providers of general-purpose AI models established by Regulation (EU) 2024/1689 (C(2025) 7719 final)Original source
  29. 2025-12-19 · announcementAnnounced Anthropic Frontier Compliance FrameworkOriginal sourceArchived copy, captured 2025-12-28
  30. 2025-12-19 · announcementEnacted New York RAISE Act (Responsible AI Safety and Education Act), as described by the GovernorOriginal source
  31. 2025-12-30 · revisionNames a commitment or law FAIF · December 2025 edition → California SB 53 (states that it complies with)Original sourceArchived copy, captured 2026-01-02
  32. 2025-12-30 · revisionIssued xAI Frontier Artificial Intelligence Framework — December 2025 editionOriginal sourceArchived copy, captured 2026-01-02
  33. 2026-01 · publicationIssued Vademecum of the Signatory Taskforce of the General-Purpose AI Code of Practice (January 2026)Original source
  34. 2026-01-01 · effectiveTakes effect California SB 53 (2025): Transparency in Frontier Artificial Intelligence Act, as chapteredOriginal source 1Archived copy, captured 2026-01-30Original source 2Archived copy, captured 2026-01-30Original source 3
  35. 2026-02 · revisionNames a commitment or law Microsoft FGF · Feb 2026 → California SB 53 (states that it responds to)Original sourceArchived copy, captured 2026-04-24
  36. 2026-02 · revisionNames a commitment or law Microsoft FGF · Feb 2026 → EU GPAI Code of Practice (states that it responds to)Original sourceArchived copy, captured 2026-04-24
  37. 2026-02 · revisionNames a commitment or law Microsoft FGF · Feb 2026 → Seoul commitments (2024) (states that it responds to)Original sourceArchived copy, captured 2026-04-24
  38. 2026-02 · revisionIssued Microsoft Frontier Governance Framework — February 2026Original sourceArchived copy, captured 2026-04-24
  39. 2026-02-24 · effectiveIssued Anthropic Responsible Scaling Policy — v3.0Original source 1Archived copy, captured 2026-02-24Original source 2Archived copy, captured 2026-02-24
  40. 2026-02-24 · effectiveNames a commitment or law RSP · v3.0 → California SB 53 (names)Original sourceArchived copy, captured 2026-02-24
  41. 2026-03-12 · publicationIssued Draft Commission Implementing Regulation on detailed arrangements for the conduct of certain proceedings by the Commission pursuant to Regulation (EU) 2024/1689 (Ares(2026)2709234)Original source 1Original source 2
  42. 2026-03-27 · signatureEnacted New York Senate Bill S8828 / Assembly Bill A9449 (2026): RAISE Act chapter amendment, signed as Chapter 96Original source
  43. 2026-04-08 · publicationIssued Meta Advanced AI Scaling Framework — Version 2Original sourceArchived copy, captured 2026-04-08
  44. 2026-05-28 · publicationNames a commitment or law FGF · May 2026 edition → California SB 53 (states that it serves as its framework under)Original sourceArchived copy, captured 2026-05-29
  45. 2026-05-28 · publicationNames a commitment or law FGF · May 2026 edition → EU GPAI Code of Practice (states that it serves as its framework under)Original sourceArchived copy, captured 2026-05-29
  46. 2026-05-28 · publicationIssued OpenAI Frontier Governance Framework — May 2026 editionOriginal source 1Archived copy, captured 2026-05-29Original source 2Archived copy, captured 2026-05-28
  47. 2026-06-30 · effectiveNames a commitment or law FAIF · June 2026 edition → EU GPAI Code of Practice (names)Original sourceArchived copy, captured 2026-07-16
  48. 2026-06-30 · effectiveIssued xAI Frontier Artificial Intelligence Framework — June 2026 editionOriginal sourceArchived copy, captured 2026-07-16
  49. 2026-07-08 · signatureEnacted Regulation (EU) 2026/1744 (Digital Omnibus on AI), as published in the Official JournalOriginal sourceArchived copy, captured 2026-07-28
  50. 2026-07-27 · effectiveTakes effect Regulation (EU) 2026/1744 (Digital Omnibus on AI), as published in the Official JournalOriginal sourceArchived copy, captured 2026-07-28
  51. 2027-01-01 · effectiveTakes effect New York Senate Bill S8828 / Assembly Bill A9449 (2026): RAISE Act chapter amendment, signed as Chapter 96Original source

Developers’ own frameworks

The earliest edition recorded here of each developer’s safety framework. Selekoa has not established that each is the first one its developer published.

Earliest recorded edition of each framework (7)

Shared external commitments, 2023–2024

The White House texts of July and September 2023 and the Frontier AI Safety Commitments of May 2024, with the listings of the developers whose frameworks are recorded here and a selection of quoted provisions.

Texts issued (3)

Developers listed as signatories (11)

Quoted provisions (11)

Where commitments and frameworks name each other

Provisions that ask for a framework, and framework editions that name the shared commitments in their own words. A framework’s statement about a commitment is the framework’s own claim.

Texts that ask for a framework (3)

Framework editions that name the shared commitments (10)

  • Qualification. The edition’s own words, recorded as naming the commitments: it describes overlap, not fulfilment. It is not evidence that the commitments are met.

  • Qualification. The edition’s own statement, quoted from it. “Helps satisfy” is recorded as a claim of fulfilment in part; it is not evidence that the commitments are met. The same sentence also names the White House commitments. RSP v3.0 and later editions contain no such sentence; that absence is not recorded as a claim.

  • Qualification. The edition’s own statement, quoted from it. “Helps satisfy” is recorded as a claim of fulfilment in part; it is not evidence that the commitments are met. The same sentence also names the Frontier AI Safety Commitments. RSP v3.0 and later editions contain no such sentence; that absence is not recorded as a claim.

  • Qualification. The edition’s own words, recorded as a response to the commitments. It is not evidence that the commitments are met.

  • Qualification. The edition’s own words, recorded as a response to the commitments. It is not evidence that the commitments are met.

  • Qualification. The edition’s own statement, quoted from it. “Helps satisfy” is recorded as a claim of fulfilment in part; it is not evidence that the commitments are met. The same sentence also names the White House commitments. RSP v3.0 and later editions contain no such sentence; that absence is not recorded as a claim.

  • Qualification. The edition’s own statement, quoted from it. “Helps satisfy” is recorded as a claim of fulfilment in part; it is not evidence that the commitments are met. The same sentence also names the Frontier AI Safety Commitments. RSP v3.0 and later editions contain no such sentence; that absence is not recorded as a claim.

  • Qualification. The edition’s own statement, quoted from it. “Helps satisfy” is recorded as a claim of fulfilment in part; it is not evidence that the commitments are met. The same sentence also names the White House commitments. RSP v3.0 and later editions contain no such sentence; that absence is not recorded as a claim.

  • Qualification. The edition’s own statement, quoted from it. “Helps satisfy” is recorded as a claim of fulfilment in part; it is not evidence that the commitments are met. The same sentence also names the Frontier AI Safety Commitments. RSP v3.0 and later editions contain no such sentence; that absence is not recorded as a claim.

  • Qualification. The edition’s own words, recorded as a response to the commitments. It is not evidence that the commitments are met.

Laws, guidance and codes: EU, California, New York and China

The EU AI Act is a regulation, adopted in 2024 and amended in 2026. The Commission’s guidelines and template explain how it reads the Act and are not binding. California SB 53 is a statute, enacted in 2025. New York’s RAISE Act is recorded through the Governor’s announcements and through the legislature’s record of its 2026 amendment, Chapter 96. China’s Interim Measures on generative AI services date from 2023. Organisations sign the EU Code, a voluntary text whose own wording says that adherence is not conclusive evidence of compliance with the AI Act. Provisions are quoted from each text, which states duties and dates. A duty or a date of application is not evidence of compliance or enforcement.

EU AI Act: adoption (1)

EU AI Act: definitions, classification and duties of general-purpose AI model providers (9)

EU AI Act: codes of practice, fines and dates of application (8)

EU Digital Omnibus on AI (2026): amendments to the AI Act (5)

EU Commission guidelines on general-purpose AI models (2025) (9)

  • Qualification. The Communication asks for approval of the draft’s content. It does not itself record that the Commission approved it. The guidelines text is a separate record (dated 19 November 2025). The statement that the guidelines apply only from formal adoption is the Communication’s; when and how adoption took place was not established beyond the later dated file.

  • Qualification. Records the dated file inspected. The library entry was last updated on 20 April 2026 and gives a publication date of 18 July 2025. Whether the November text differs from the draft content approved in July 2025 was not compared, so provisions are dated by the November file only.

  • Qualification. The Commission’s statement of its planned approach, in a document of November 2025. Dates of application are recorded separately from Article 113. No enforcement action under the Regulation is recorded; a missing record is a gap in coverage, not a finding that none occurred. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. An indicative criterion that the Commission says it uses; it is not a threshold in the Act, whose definition in Article 3(63) governs. The guidelines say the Commission’s approach may change as technology and the market evolve (paragraph 16). The exponent is a superscript in the PDF, and the text layer renders 10²³ as 1023. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. The Commission’s interpretation of Article 52(1); the two-week deadline is in the Regulation (a separate record). Not evidence that any notification was made, or by whom. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. The Commission’s stated expectation. The second part is introduced with “may” and gives the reason that the AI Office would have less understanding of such providers’ compliance. No provider is named and no request is recorded. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. The Commission’s stated approach. The Code’s own text, in a separate record, says that adherence is not conclusive evidence of compliance; the two propositions differ and neither is a finding about any provider. The guidelines describe the code as assessed as adequate by the AI Office and the Board; the Commission’s opinion of 1 August 2025 is recorded elsewhere. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. A discretionary statement of approach ("may"). It cites Article 101(1); the Article’s fine ceiling is recorded separately and its own text was not searched for this point. No fine is recorded. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. A statement about the status of the guidelines, in the guidelines themselves. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

EU template for the public summary of training content (2)

  • Qualification. Section 2.3 of the Template as it appears in the Communication. The obligation to publish a summary comes from Article 53(1)(d) of the AI Act. No summary published by any provider was inspected. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. Paragraph (29) of the Explanatory Notice. The PDF misspells "intervals" as "intervanls"; the paraphrase corrects the spelling. The Notice also states that summaries of models placed on the market before 2 August 2025 are due by 2 August 2027; that sentence is not recorded here. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

EU AI Office procedure: a draft implementing regulation (2026) (2)

  • Qualification. A draft, not an adopted act. The feedback record, read on 2026-10-11 and last modified on 2026-06-26, gives a feedback period ending 2026-04-09, lists the feedback status as closed and the adoption stage as upcoming, with no adoption date; whether an act was adopted afterwards was not checked in the Official Journal. The draft’s views are, in its own words, preliminary views of the Commission services.

  • Qualification. What the draft proposes, not law. It would change if adopted in another form. The draft says the Commission may require a provider to disable logging that could record the Commission’s access; that sentence is not recorded here. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

EU Code of Practice: issue, signatures, a refusal and the Signatory Taskforce (12)

EU Code of Practice: quoted provisions (Safety and Security, Transparency and Copyright chapters) (8)

  • Qualification. What the text says; whether a given model has systemic risk is a question of the AI Act’s classification and is not recorded. Quoted from the Safety and Security chapter.

  • Qualification. What the text says. Quoted from the Safety and Security chapter.

  • Qualification. A commitment that signatories make, subject to the intellectual-property and confidentiality conditions that the measure states. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. A commitment that signatories make; the Commission’s signatory page lists xAI as signed to the Safety and Security chapter only (a separate record), so the Copyright chapter is not recorded as applying to it. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. What the text says, not evidence that any incident was reported. Quoted from the Safety and Security chapter.

  • Qualification. What the text says; acceptability is determined under the chapter’s own Measure 4.1. Quoted from the Safety and Security chapter.

  • Qualification. A commitment that signatories make; the chapter’s objectives say that adherence is not conclusive evidence of compliance (separate record). The chapter and the Model Documentation Form are not summarised here. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. What the text says. Quoted from the Safety and Security chapter.

EU Code of Practice: framework editions that name it (3)

  • Qualification. The edition’s own words, recorded as a response to the Code; the same change-log sentence is the edition’s link to California SB 53. It is not evidence of compliance.

  • Qualification. The edition’s own designation, quoted from it; it presents itself as the public summary of the framework the Code calls for, not as that framework in full. It is not evidence of compliance. The same edition names OpenAI Ireland Limited as the provider responsible for the Code in the EU; that statement is not recorded.

  • Qualification. The edition’s own words, recorded as naming the chapter as the basis of its risk domains. It is not a statement that the framework complies with the Code, and it is not evidence of compliance.

California SB 53: enactment (1)

California SB 53: quoted provisions (18)

California SB 53: documents that name it (5)

  • Qualification. An announcement by Anthropic about its own document; the framework’s own text, version and dates were not inspected in this pass, so no record links the framework itself to SB 53. The stated purpose is forward-looking (“will serve”). It is not evidence that the framework satisfies the statute or that Anthropic is a large frontier developer under it. Anthropic makes this statement in its announcement, not in the FCF itself, which was not retrieved. A designation is not evidence that the FCF satisfies § 22757.12 or that Anthropic complies. The same post says Anthropic endorsed SB 53 (its stated position, not recorded here). For the RSP side of the split, see the record of RSP v3.0’s reference to SB 53.

  • Qualification. The edition’s own statement about SB 53, quoted from it. It is not evidence that the edition satisfies the statute, that its publisher is a frontier developer under it, or of implementation or compliance. This is xAI's own assertion that the FAIF document complies with TFAIA. It is not evidence of compliance. Whether the FAIF satisfies the Act, and whether and how the Act applies to xAI, are legal questions left to the stop list. The edition's footnote 1 quotes TFAIA's catastrophic-risk definition. This record complements the existing record of the June 2026 edition’s changes, which notes that the June 2026 edition lacks the sentence, and does not duplicate it. The evidence for that record already quotes the sentence's opening words from the same PDF, so the overlap counts against this document's quotation budget.

  • Qualification. The edition’s own statement about SB 53, quoted from it. It is not evidence that the edition satisfies the statute, that its publisher is a frontier developer under it, or of implementation or compliance. The inspected passages do not say that this document is Microsoft's TFAIA frontier AI framework, or that Microsoft is a large frontier developer; neither is inferred. The change log also lists aligning the update cadence with regulatory obligations and aligning the framework's scope with that of regulations. It does not say which regulation drives which change.

  • Qualification. The edition’s own statement about SB 53, quoted from it. It is not evidence that the edition satisfies the statute, that its publisher is a frontier developer under it, or of implementation or compliance. This is an explicit link that separates the RSP from Anthropic's SB 53 compliance documents. The footnote does not name the Frontier Compliance Framework or any other specific document. Paraphrase of the Introduction's body: the RSP may serve some regulatory requirements but is not designed to be comprehensive, and regulatory requirements that exceed or differ from it will be addressed through separate documents. On 2026-10-04 a whitespace-normalised text search of the live PDFs found the same footnote sentence once each in v3.1, v3.2, v3.3 and v3.4. This record covers v3.0 only; no per-edition records are created here. The record makes no claim about whether any compliance document satisfies SB 53.

  • Qualification. The edition’s own statement about SB 53, quoted from it. It is not evidence that the edition satisfies the statute, that its publisher is a frontier developer under it, or of implementation or compliance. This is OpenAI's own designation and is not evidence of compliance or adequacy. The same introduction presents the FGF as a public summary of OpenAI's Safety & Security Framework under the EU GPAI Code of Practice (an EU link outside this topic). The existing record of OpenAI’s description of the FGF as applying relevant parts of its Preparedness approach covers the relation to the Preparedness Framework. The FGF is the only OpenAI framework edition in the dataset that names TFAIA. What document, if any, OpenAI designated under TFAIA between 2026-01-01 and 2026-05-28 was not established.

New York RAISE Act: signing announced, and Chapter 96 (2026) (6)

  • Qualification. An announcement by the Governor’s office about a signing. The legislature’s bill record for S8828 lists a chapter amendment signed on 27 March 2026 as Chapter 96, and gives S.6953-B and A.6453-B as the bill "as proposed" whose chapter it amends, so the bill numbers in the release belong to the original bill. The date and chapter number of that original signing were not located in an official record; secondary summaries report December 2025. The penalty figures and the 72-hour reporting window in the release’s context evidence match the text of S8828 as inspected on 2026-10-11 (separate records). The Governor’s release of 21 September 2026 says compliance is required beginning January 2027, the effective date in S8828. Signing is not evidence of application to a named developer, compliance or enforcement.

  • Qualification. The legislature’s record of the signing; the Governor’s message and the chapter in the session laws were not inspected. The relation to the announcement of 19 December 2025 (a separate record) is as the bill describes it: it amends a chapter of 2025 "as proposed in" S.6953-B and A.6453-B. The signing date and chapter number of that 2025 chapter were not located in an official record. Signing is not evidence of application to a named developer, compliance or enforcement.

  • Qualification. A ceiling and a power, not a record of any penalty. The bill also states that it creates no private right of action. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. The bill defines "critical safety incident" and "office" (an office within the Department of Financial Services); neither definition is reproduced here. A 24-hour disclosure duty for imminent risk of death or serious physical injury is in the next paragraph and is not recorded. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. A large frontier developer is, in the bill’s definition, a frontier developer with annual gross revenues above five hundred million dollars together with its affiliates; a frontier model is a foundation model trained with computing power above a stated quantity. Whether any named company falls within these definitions is legal application and is not recorded. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. Read from the bill as printed on 8 January 2026, which the action list does not show as amended; the chaptered text was not inspected. The text shows the replaced wording in brackets. Section 5 of the bill makes the bill itself effective immediately, except that sections one, two and four take effect with the 2025 chapter. A date of effect is not evidence that any developer is subject to the law or complies. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

China: Interim Measures on generative AI services (2023) (3)

  • Qualification. Records the order as posted on the issuing authority’s site, in Chinese; the English statement is a translation by a machine agent. Later amendments, if any, were not checked. Issuing a measure is not evidence of its application to a named developer, compliance or enforcement.

  • Qualification. English rendering by a machine agent from the Chinese text, which controls. The definition of "generative AI services" and the exclusions that follow in Article 2 are not recorded. Whether any named developer provides such a service is not recorded. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

  • Qualification. English rendering by a machine agent from the Chinese text, which controls; the rendering of “舆论属性或者社会动员能力” is a translation choice. The text does not say which services have these attributes, and no record shows that any frontier developer’s service was assessed or filed. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.

Framework revisions in the same period

Key revisions only: each publisher’s new major version or new framework, or the later edition where editions carry no version number. A revision recorded near a commitment’s date is not recorded as a response to it.

Key revisions and what they replace (15)

Recorded changes from the previous edition (30)

Provisions not found in the inspected text (16)

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent. Text comparison only; no replacement between the editions is asserted by this record.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent. Text comparison only; no replacement between the editions is asserted by this record.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent. Text comparison only; no replacement between the editions is asserted by this record.

  • Qualification. A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent. Text comparison only; no replacement between the editions is asserted by this record.

Open questions and unresolved records

Records marked contested or unresolved, followed by open questions and questions set aside on the commitments and document histories pages that concern a record listed here.

Contested and unresolved records (4)

Open questions (17)

  • The White House announced on 26 July 2024 that Apple had signed onto the voluntary commitments. The announcement does not say whether Apple took up the July or the September 2023 text, so no record links Apple to either.

    Records: ·

  • The September 2023 text measures which models are in scope against each company’s own most advanced model, where the July text uses the industry frontier. The September text does not say whether its clause also applies to the seven companies of July 2023.

    Records:

  • The GOV.UK text lists “Google”, while the UK government’s press release of the same day lists “Google / Google DeepMind”. Neither says which legal entity agreed.

    Records:

  • The commitments ask for a safety framework by “the upcoming AI Summit in France” without giving a date; the UK government’s press release of the same day places that summit in early 2025. Whether each organisation published a framework in time is not recorded.

    Records:

  • The Commission’s list does not say which chapters each organisation signed. The exception is xAI: the Commission says it signed up to the Safety and Security chapter and must show compliance on transparency and copyright by other means. The provisions recorded here are quoted from that chapter.

    Records: ·

  • Of the 26 organisations listed on 1 August 2025, eight are absent from the page as last updated on 31 July 2026, which adds three others. The Commission gives no reasons; its questions and answers say a signatory may withdraw at any time. Every organisation recorded here is still listed.

    Records:

  • Anthropic announced its Frontier Compliance Framework on 19 December 2025 as its compliance framework for SB 53. The framework is served from a portal that loads its content by script, and its own text was not inspected for this page. No record therefore links the framework itself to SB 53. An archived copy has since been found and awaits verification.

    Records:

  • Responsible Scaling Policy v3.0 names SB 53 only to say that statutory requirements with specific thresholds are addressed in separate compliance documentation. It does not present itself as Anthropic’s framework under the statute.

    Records: ·

  • xAI’s December 2025 framework states that it complies with the statute; the June 2026 edition does not contain that sentence. Neither the sentence nor its absence establishes compliance or non-compliance.

    Records: ·

  • No OpenAI text inspected so far names the December 2023 Beta as the edition that Version 2 replaces, so the replacement record is unresolved and carries context evidence only. A separate question is contested: whether Version 2’s governance arrangements let OpenAI Leadership deploy High-capability models without sufficient safeguards. The two sides are an academic analysis of leadership discretion and OpenAI’s stated predeployment requirement. The rule’s text itself is recorded as a supported provision.

    Records: ·

  • Two different texts carry the Version 1.1 label: the earliest archive capture (2025-02-13) and a revised wording captured from 2025-03-28 onwards. The provisions quote the earlier text. No announcement of the revision was found.

    Records:

  • The February 2026 edition is identified by its version month. Its publication date is not established. The file was last modified on 2026-03-24 and first captured on 2026-04-24.

    Records:

  • Two questions remain open. Did the v2.0 revision make Anthropic’s commitments less verifiable? Do v3.0’s roadmap and transparency mechanisms compensate for its reduced unilateral commitments? Each record keeps the published assessments as context, because none directly opposes another on the same question. The changes to the text are recorded separately.

    Records: ·

  • No record links the June 2026 edition to the December 2025 edition as its replacement, because no explicit statement that one supersedes the other was found. The one recorded difference was inferred by comparing the two documents. It is not based on a statement by xAI.

    Records: · ·

  • The December 2025 edition is identified by its printed “last updated” date. The file name carries a different day, as the date note records. Neither edition has an established publication day of its own.

    Records: ·

  • The Frontier Governance Framework is recorded as applying relevant parts of the continuing Preparedness approach. Its announcement names no Preparedness version. The link therefore points to the Preparedness framework as a whole rather than to Version 2, and the new framework is not recorded as its successor.

    Records:

  • Most Responsible Scaling Policy revisions are identified by their effective dates. Where no publication day was established, none is inferred.

    Records: · · · · · ·

Questions set aside (9)

These questions were deliberately not turned into records. Answering them would need legal interpretation, a decisive source for conflicting dates, or statements about a living person beyond documented roles.

  • Needs legal interpretation Whether the 2023 White House commitments bind or ever bound the companies, and whether they remain in effect. The July text calls them voluntary and says the companies intend them to remain in effect until regulations covering substantially the same issues come into force. Related records:
  • Needs legal interpretation Whether being listed for the Frontier AI Safety Commitments creates any obligation, and whether “agreed to”, “signed up to” and “endorse” (the wording of the text, of the UK press release and of company statements quoted in it) describe the same act. Related records:
  • Needs legal interpretation The legal effect of signing or adhering to the EU Code of Practice, the consequence of xAI’s chapter-limited signature, and whether the Code goes beyond the AI Act, as Meta said in refusing to sign. The Commission’s opinion concludes that the Code adequately covers the obligations in Articles 53 and 55 of the Act; that is a different question. Related records: · ·
  • Needs legal interpretation Whether any named company is a provider of a general-purpose AI model with systemic risk under the AI Act, to which the Safety and Security chapter is addressed. Related records:
  • Needs legal interpretation Whether SB 53 applies to a named company is a question of legal application, for example whether OpenAI, Google, Anthropic or xAI is a “frontier developer” or a “large frontier developer”. A competent secondary reference (METR, January 2026) states that such developers are governed by the statute. That statement is not recorded as a finding. Related records: ·
  • Needs legal interpretation An advocacy organisation has alleged that OpenAI violated SB 53 in connection with model releases in 2026; OpenAI disputes this, partly by contesting how its own framework wording should be read. Whether any violation occurred is a question of legal effect and compliance and is not recorded. Related records: ·
  • Needs legal interpretation Company designations are recorded as the companies’ own statements, never as legal fact: OpenAI’s designation of its Frontier Governance Framework as its framework under the Act (and of an OpenAI operating company as responsible for compliance), xAI’s statement that its December 2025 framework complies, and Anthropic’s designation of its Frontier Compliance Framework. Related records: · ·
  • Needs legal interpretation The Act’s carve-outs for strict conflicts with federal contracts, federal preemption and local regulation (SEC. 5) and the federal-equivalence pathway’s effect on obligations require legal interpretation and are not recorded beyond the quoted text. Related records:
  • Needs legal interpretation What it means to comply with one’s own framework when the framework uses discretionary or hedged language, and what counts as a material modification or a substantially modified model, are questions of interpretation and are not recorded. Related records: · ·

What Selekoa cannot answer here

Questions these records leave open. Each is a gap in the records, not a finding about what happened.

  • Implementation. Whether any developer did what its framework or a commitment says. No record here states that a framework threshold was assessed, reached or acted on for a particular model.

  • Compliance. Whether any developer complied with a commitment, the EU Code or SB 53. Where a framework says that it fulfils, complies with or serves as a framework under a text, that is the framework’s own statement. The Commission’s guidelines say how adherence to the EU Code bears on its enforcement approach. No statement about a named developer is recorded.

    Records: · · · · ·

  • Enforcement. Whether SB 53 or the EU AI Act has been enforced against any developer. SB 53’s penalty and enforcement provisions and the AI Act’s fine provision for providers of general-purpose AI models, with its dates of application, are recorded. No enforcement action is. The Commission’s guidelines state that its enforcement powers apply from 2 August 2026, and a draft implementing regulation on evaluations and proceedings was published for feedback in March 2026 and is not recorded as adopted. An alleged SB 53 violation is listed above as a question set aside.

    Records: · · · · ·

  • Law. Parts of the EU AI Act that are not recorded. Selekoa quotes selected provisions of the Act as published in the Official Journal in 2024 and three amendments made in 2026. It does not hold the consolidated text, the annexes (including the documentation items in Annex XI) or the full duties in Article 53(1) on downstream information, copyright and training-data summaries. Commission guidance is recorded in part: the guidelines of 2025, the training-content template and the Transparency and Copyright chapters of the Code, each with selected provisions. The Commission’s questions-and-answers pages and the remaining guidance are not.

    Records: · · ·

  • Other jurisdictions. Law and regulation beyond California SB 53, the EU AI Act, the EU Code and the New York RAISE Act. The New York RAISE Act is recorded through two announcements by the Governor and through the legislature’s printed bill for its 2026 amendment (Chapter 96), from which provisions are quoted. The 2025 chapter’s own text and the session-law print were not inspected. China’s Interim Measures on generative AI services (2023) are recorded in part, in Chinese. No United Kingdom text is recorded: a search on 2026-10-11 found no bill text. No other statute, regulation or official code on frontier AI developers is recorded. Enacted laws recorded in Selekoa: California SB 53; EU AI Act (2024); EU Digital Omnibus on AI (2026); New York RAISE Act; New York RAISE Act, Chapter 96 (2026); China generative AI interim measures (2023).

    Records: · · · · ·

  • Reception. How governments, researchers, journalists or developers’ staff responded to these texts. Published commentary appears only as evidence in the contested and unresolved records above.

    Records: · ·

  • Causes of change. Whether a commitment or a law led a developer to revise its framework. Only links that a framework states in its own words are recorded. Closeness in date never creates a link.

    Records: ·

  • Coverage. Links not yet recorded. A missing link is a gap in coverage. The framework may still name one of these texts. Frameworks with no edition recorded as naming any of these texts: Google DeepMind · Frontier Safety Framework; OpenAI · Preparedness Framework.

    Records: · · · · ·

  • Coverage. Frameworks of other signatories. The Frontier AI Safety Commitments ask each listed organisation to publish a safety framework. Whether these organisations did so is not recorded. Organisations listed as signatories of Seoul commitments (2024) with no framework recorded here: 01.AI, Amazon, Cohere, G42, IBM, Inflection AI, Magic, MiniMax, Mistral AI, Naver, NVIDIA, Samsung Electronics, Technology Innovation Institute, Zhipu AI.

    Records: · · · · · · · · · · · · · ·

  • Coverage. Whether the earliest recorded edition of each framework is the first its developer published. Earlier drafts or differently named documents may exist.

  • Documents not inspected. Anthropic’s Frontier Compliance Framework is recorded only through its announcement. Which document, if any, OpenAI designated under SB 53 between 2026-01-01 and 2026-05-28 has not been established. Attempts on 2026-10-10 and 2026-10-11 found that the portal that serves the Frontier Compliance Framework loads its content by script, that archived captures of its pages hold only the page shell, and that a forum comment of January 2026 describes the framework as page images that cannot be copied. Searches on 2026-10-11 of OpenAI’s published pages found no statement that designates a document under SB 53 before 28 May 2026. The only dated designation recorded is the Frontier Governance Framework’s own statement.

    Records: ·