At a glance

  • A model card is a voluntary format from a 2019 research paper, but California SB 53 and, from January 1, 2027, New York’s RAISE Act and Illinois’s AI Safety Measures Act deem a frontier developer in compliance with the transparency report duty if it publishes the required information as part of a larger document, including a system card or model card.
  • Under the EU AI Act, general-purpose AI model providers have owed Annex XI documentation to the AI Office and national competent authorities on request, and Annex XII information to providers of AI systems that intend to integrate the model, since August 2, 2025 (August 2, 2027 for models placed on the market before then). An exception for models released under a free and open-source licence with their weights, architecture, and usage information made public covers both duties, but not models with systemic risk. High-risk system providers owe Annex IV documentation and Article 13 instructions for use from December 2, 2027 (Annex III) or August 2, 2028 (Annex I Section A).
  • ONC’s HTI-5 final rule, published October 28, 2026, removes the health IT decision support source attribute (“model card”) and intervention risk management requirements in 45 CFR 170.315(b)(11)(iv) to (vi), effective December 28, 2026; until then they still apply. FDA’s example model card is in a draft guidance, and FDA says it does not require a model card or a specific model card format.

What is an AI model card?

An AI model card is a short document published with a trained machine learning model that states what the model is for, how it was evaluated, how it performs across relevant groups and conditions, and where it should not be used. The term comes from “Model Cards for Model Reporting” by Margaret Mitchell and eight co-authors, presented at FAT* ’19 in January 2019 (arXiv 1810.03993). It is a voluntary practice, not a legal term, though three state laws accept a model card as the container for a required disclosure.

The paper proposes nine sections: model details, intended use, factors, metrics, evaluation data, training data, quantitative analyses, ethical considerations, and caveats and recommendations. Its authors say the sections are “not intended to be complete or exhaustive.” The paper puts its weight on disaggregated evaluation: performance reported across groups, such as race, sex, or Fitzpatrick skin type, and intersections such as age and race, not one overall score. A statute asks different questions: who must document, what each recipient must receive, and when.

Which laws require model documentation?

As of November 3, 2026, binding duties come from the EU AI Act, California SB 53 and AB 2013, ONC’s health IT certification criteria until December 28, 2026, and, from January 1, 2027, New York’s RAISE Act, Illinois’s AI Safety Measures Act, and Colorado SB 26-189. FDA’s draft guidance, the GPAI Code of Practice, and the NIST AI RMF are voluntary.

Model documentation duties and frameworks, as of November 3, 2026
Law or frameworkWho must documentWhat must be documentedTo whomWhen
EU AI Act, Article 11 and Annex IV (technical documentation)Providers of high-risk AI systemsAt a minimum the Annex IV elements: general description, development process and design choices, data requirements, human oversight assessment, validation and testing, monitoring and performance, risk management system, changes, standards applied, declaration of conformity, post-market monitoring planNational competent authorities and notified bodies; kept at the disposal of national competent authorities for 10 years (Article 18)Drawn up before the system is placed on the market or put into service and kept up to date; applies from December 2, 2027 (Annex III) or August 2, 2028 (Annex I Section A)
EU AI Act, Article 13 (instructions for use)Providers of high-risk AI systemsAt least: provider identity; intended purpose; accuracy, robustness, and cybersecurity levels; foreseeable risks; output explanation capabilities; performance on specific persons or groups when appropriate; input data specifications; information to interpret output; pre-determined changes; human oversight measures; resources, lifetime, and maintenance; loggingDeployersAccompanies the system; same application dates as Article 11
EU AI Act, Article 53(1)(a) and Annex XIProviders of general-purpose AI modelsTechnical documentation of the model, including its training and testing process and the results of its evaluation; Section 2 adds evaluation strategies, adversarial testing, and architecture for models with systemic riskAI Office and national competent authorities, on requestSince August 2, 2025; by August 2, 2027 for models placed on the market before August 2, 2025
EU AI Act, Article 53(1)(b) and Annex XIIProviders of general-purpose AI modelsInformation that enables downstream providers to understand the model’s capabilities and limitations and to comply with their own obligationsProviders of AI systems that intend to integrate the modelSame as Annex XI
GPAI Code of Practice, Transparency chapter, Model Documentation Form (voluntary)Signatory providers of general-purpose AI modelsOne form covering Annex XI Section 1 and Annex XII, with each item marked for the AI Office, national authorities, or downstream providersDownstream providers; the AI Office on requestWhen the model is placed on the market; prior versions kept 10 years
California SB 53, Bus. & Prof. Code 22757.12(c)Frontier developers (foundation model trained with more than 10²⁶ operations); added content for large frontier developers (with affiliates, more than $500 million in annual gross revenues in the preceding calendar year)Transparency report: website, contact mechanism, release date, languages, output modalities, intended uses, use restrictions; large developers add catastrophic risk assessment summariesThe public, on the developer’s websiteBefore, or concurrently with, deploying a new or substantially modified frontier model; in effect since January 1, 2026
California AB 2013, Civil Code 3111Developers of generative AI systems or services released on or after January 1, 2022 and made publicly available to Californians, with the exemptions in 3111(b)A high-level summary of the training datasets, with 12 listed itemsThe public, on the developer’s websiteOn or before January 1, 2026, and before each later release or substantial modification
New York RAISE Act, Gen. Bus. Law Article 44-B, § 1421(3)Frontier developers, for frontier models developed, deployed, or operating in whole or in part in New York; added content for large frontier developersTransparency report with the same seven items as SB 53, plus catastrophic risk summaries for large developersThe public, on the developer’s websiteFrom January 1, 2027
Illinois AI Safety Measures Act, Public Act 104-0538, Section 10(c)Frontier developers (foundation model trained with more than 10²⁶ operations, counting fine-tuning, reinforcement learning, and other material modifications); added content for large frontier developers (with affiliates, more than $500 million in annual gross revenues in the preceding calendar year)Transparency report with the same seven items as SB 53, plus catastrophic risk summaries for large developers, provided in a machine-readable formatThe public, on the developer’s websiteBefore, or concurrently with, deploying a new or substantially modified frontier model; from January 1, 2027
Colorado SB 26-189, C.R.S. 6-1-1702Developers of covered automated decision-making technologyIntended and known harmful uses, training data categories, known limitations and risks, instructions for use, monitoring, and human review, and what deployers need for their own disclosures; compliance records kept at least three yearsDeployersOn and after January 1, 2027, with notice of material updates; a court order currently bars Attorney General enforcement (see the Colorado section)
ONC HTI-1, 45 CFR 170.315(b)(11)(iv) to (vi)Developers of certified health IT that supply decision support interventionsSource attributes for evidence-based and predictive interventions, and intervention risk managementA limited set of identified users of the certified health ITIn effect until December 28, 2026, when the HTI-5 final rule removes it
FDA draft guidance on AI-enabled device software functions (non-binding)Sponsors of AI-enabled devicesExample model card in Appendix E; example 510(k) summary with model card in Appendix FUsers and the public, through labeling and the public submission summaryDraft issued January 2025; not final as of November 3, 2026
NIST AI RMF 1.0 and AI 600-1 (voluntary)Any organization that chooses to use themKnowledge limits and human use of output (MAP 2.2); model explained, validated, and documented (MEASURE 2.9)Internal; no filingAcross the lifecycle

What does the EU AI Act require for high-risk AI systems?

The Act asks for two documents with different readers. Article 11(1) requires the provider to draw up technical documentation before a high-risk system is placed on the market or put into service and to keep it up to date. It must demonstrate that the system complies with the requirements of Chapter III, Section 2 and give national competent authorities and notified bodies the necessary information in a clear and comprehensive form to assess that compliance, and it must contain, at a minimum, the elements in Annex IV. Article 18 requires the provider to keep it at the disposal of national competent authorities for 10 years after the system is placed on the market or put into service. As amended by Regulation (EU) 2026/1744, Article 11(1) lets SMEs, including start-ups, and small mid-cap enterprises provide the Annex IV elements in a simplified manner, on a form the Commission must establish.

Article 13 is the deployer’s document: instructions for use with “concise, complete, correct and clear information that is relevant, accessible and comprehensible to deployers.” Two of its Article 13(3) items read most like a model card: the level of accuracy, including its metrics, robustness, and cybersecurity referred to in Article 15 against which the system has been tested and validated and which can be expected, and any known and foreseeable circumstances that may have an impact on that expected level of accuracy, robustness, and cybersecurity; and, when appropriate, the system’s performance regarding specific persons or groups of persons on which it is intended to be used.

Under amended Article 113(c), both articles apply from December 2, 2027 for Annex III systems and from August 2, 2028 for systems classified under Article 6(1) and Section A of Annex I. Under Article 2(2), only Article 6(1), Article 60a, and Articles 102 to 112 apply to systems related to products covered by Section B of Annex I. The EU high-risk AI systems guide covers classification and systems already on the market.

What must general-purpose AI model providers document?

Since August 2, 2025, Article 53(1) has required every provider of a general-purpose AI model to:

  • (a) draw up and keep up to date the technical documentation of the model, including its training and testing process and the results of its evaluation, containing at a minimum the information in Annex XI, for the purpose of providing it, upon request, to the AI Office and the national competent authorities
  • (b) draw up, keep up to date, and make available information and documentation to providers of AI systems who intend to integrate the model, which must enable them to have a good understanding of the model’s capabilities and limitations and to comply with their own obligations, and must contain at a minimum the elements in Annex XII, without prejudice to the need to observe and protect intellectual property rights and confidential business information or trade secrets

Providers of models placed on the market before August 2, 2025 have until August 2, 2027 (Article 111(3)). Points (c) and (d), the copyright policy and the public training content summary, are covered in the AI data governance guide.

Article 53(2) exempts from points (a) and (b) providers of models released under a free and open-source licence that allows for the access, usage, modification, and distribution of the model, and whose parameters, including the weights, the information on the model architecture, and the information on model usage, are made publicly available. The exception does not apply to general-purpose AI models with systemic risks, and it does not reach points (c) and (d). Annex XI Section 2 adds evaluation strategies, adversarial testing, and architecture for systemic-risk models.

The voluntary GPAI Code of Practice, published July 10, 2025 and found adequate by the Commission on August 1, 2025, turns the Annex XI Section 1 and Annex XII content into a Model Documentation Form. Its sections cover general information, model properties, distribution and licences, use, training process, training data, computational resources, and energy consumption, and each item is marked for the AI Office, national competent authorities, downstream providers, or a combination. Under Measure 1.1, signatories document at least the information in the form when they place a model on the market and keep previous versions for 10 years after the model is placed on the market. Using the form itself is optional. Under Measure 1.2, they publish a contact for documentation requests and answer necessary downstream requests within a reasonable timeframe, and no later than 14 days of receiving the request save for exceptional circumstances.

Does a model card satisfy California SB 53, New York’s RAISE Act, and Illinois’s AI Safety Measures Act?

Yes, if it carries the required items. Under Business and Professions Code 22757.12(c)(1), before, or concurrently with, deploying a new frontier model or a substantially modified version of an existing one, a frontier developer must clearly and conspicuously publish on its website a transparency report containing its website, a mechanism for a natural person to communicate with it, the model’s release date, supported languages, supported output modalities, intended uses, and any generally applicable restrictions or conditions on use. A large frontier developer must add summaries of its catastrophic risk assessments under its frontier AI framework, their results, the extent to which third-party evaluators were involved, and other steps taken to fulfill the requirements of the framework with respect to that model (22757.12(c)(2)).

Subdivision (c)(3) settles the format: a developer that publishes this information “as part of a larger document, including a system card or model card, shall be deemed in compliance with the applicable paragraph.” Subdivision (f) allows redactions necessary to protect the developer’s trade secrets, its cybersecurity, public safety, or the national security of the United States, or to comply with federal or state law; the developer must describe the character and justification of each redaction to the extent those concerns permit and retain the unredacted information for five years. Only the Attorney General may recover SB 53 civil penalties, which apply to large frontier developers, depend on the severity of the violation, and may not exceed $1,000,000 per violation.

New York’s RAISE Act, signed as Chapter 699 of 2025 on December 19, 2025 and amended by Chapter 96 of 2026 (A.9449) on March 27, 2026, takes effect January 1, 2027. Section 1421(3) requires the same seven-item report and large-developer summaries and accepts the same “system card or model card” container. It applies only to frontier models developed, deployed, or operating in whole or in part in New York. The Attorney General may seek civil penalties, based on the severity of the violation, of up to $1 million for a first violation and up to $3 million for each subsequent violation where a large frontier developer fails to publish or transmit a compliant document, among other listed failures.

Illinois’s Artificial Intelligence Safety Measures Act (SB 315, Public Act 104-0538), approved July 6, 2026, takes effect January 1, 2027. Section 10(c) requires every frontier developer to publish the same seven-item transparency report before, or concurrently with, deploying a new or substantially modified frontier model, requires large frontier developers to add the same four catastrophic risk summaries in a machine-readable format, and accepts the same “system card or model card” container (Section 10(c)(4)). Section 10(g) allows redactions on the same grounds as SB 53 and requires the same description of each redaction and five-year retention of the unredacted information. The Act’s other duties for large frontier developers are not model documentation: a disclosure statement filed with the Illinois Emergency Management Agency and Office of Homeland Security from January 1, 2027, and, from January 1, 2028, a published frontier AI framework and an annual independent third-party audit. Only the Attorney General may recover civil penalties under Section 25, of up to $1,000,000 for a first violation and up to $3,000,000 for each subsequent violation, where a large frontier developer fails to publish or transmit a compliant document, among other listed failures.

AB 2013 is about data, not performance: Civil Code 3111 requires a 12-item training data summary. It exempts systems whose sole purpose is to help ensure security and integrity or is the operation of aircraft in the national airspace, and systems developed for national security, military, or defense purposes that are made available only to a federal entity (3111(b)). In X.AI LLC v. Bonta, the Central District of California denied a preliminary injunction on March 4, 2026; xAI’s appeal (Ninth Circuit No. 26-1591) was pending as of November 3, 2026, with oral argument calendared for November 18, 2026, and the law is enforceable while it proceeds. The California AI laws guide covers both statutes.

What happened to the health IT “model card” requirement?

It ends on December 28, 2026. Until then, ONC’s HTI-1 rule, published January 9, 2024, requires developers of certified health IT that supply decision support interventions to support source attributes (45 CFR 170.315(b)(11)(iv)), let a limited set of identified users access and modify them ((v)), and apply intervention risk management practices ((vi)).

The HTI-5 final rule (91 FR 67916, October 28, 2026) removes paragraphs (b)(11)(iv) through (vi) and the related maintenance requirement in 170.402(b)(4), effective December 28, 2026. ONC’s preamble says the source attribute requirements support “model cards” for predictive interventions and that it identified no evidence users relied on them or that they led to positive impacts on patient care. It encourages users to ask developers for that information, reminds them of the bias duty in 45 CFR 92.210, and says it will consider in future rulemaking whether some elements should be restored. The healthcare AI guide covers Section 1557 and the state rules.

Does FDA require a model card for an AI-enabled medical device?

No. FDA’s draft guidance on AI-enabled device software functions (January 2025, docket FDA-2024-D-4488) gives an example model card in Appendix E and an example 510(k) summary with a model card in Appendix F, and states that FDA “does not require the inclusion of a model card or a specific model card format, and this example should not be considered a template.” FDA presents a model card as one way to communicate information that may align with what FDA may require, for example in the public 510(k) summary and labeling. FDA’s guidance page still labeled the document a draft, “Not for implementation,” on November 3, 2026. The FDA PCCP guide covers the change control plan guidance beside it.

What must a developer give deployers under Colorado SB 26-189?

From January 1, 2027, a developer of covered automated decision-making technology must make available to each deployer, in a form and manner reasonably understandable to a deployer that protects trade secrets and information protected from disclosure by law, the items in the table above, with notice if any information is withheld (C.R.S. 6-1-1702(1)), and must notify deployers of material updates within a reasonable time (6-1-1702(2)). The duties reach only a deployer’s use of covered technology that was marketed, advertised, configured, contracted, sold, or licensed to be used to materially influence a consequential decision (6-1-1702(3)). The developer must also keep, for at least three years after a record is created, records reasonably necessary to demonstrate compliance, including system version identifiers, changelogs, and the documentation and update notices it gave deployers (6-1-1702(4)). A federal court order of April 27, 2026 in X.AI LLC v. Weiser bars the Colorado Attorney General from initiating enforcement, including investigations, for violations occurring on or before 14 days after the court rules on xAI’s preliminary injunction motion, which xAI must file within 28 days after final adoption of rules implementing the law; the public docket showed no such motion as of November 3, 2026. The Colorado SB 26-189 guide covers enforcement.

Where do the NIST AI RMF and ISO/IEC 42001 fit?

In the voluntary NIST AI RMF 1.0, MAP 2.2 asks that information about the system’s knowledge limits and how its output may be utilized and overseen by humans be documented, and MEASURE 2.9 asks that the model be explained, validated, and documented. The Generative AI Profile, NIST AI 600-1, suggests reviewing transparency artifacts, such as system cards and model cards, for third-party models (MG-3.1-005). ISO/IEC 42001 is also voluntary. This guide does not map its Annex A controls to model card fields because we could not verify the control numbers from a primary source. The NIST AI RMF and AI governance frameworks guides cover both.

Model card template: which fields map to which law?

This outline is a starting point and is not complete for any one law. Each field is tagged with the provisions it helps satisfy; a tag does not mean the law reaches your model. It leaves out, among other things, Annex IV points 5, 7, 8, and 9 (risk management system, standards applied, declaration of conformity, post-market monitoring plan), Annex XI Section 2, most of AB 2013’s 12 items, and the Model Documentation Form’s authenticity and measurement-methodology fields. Check each applicable provision in full.

  1. Identity: name, version, release date, developer, contact. (Mitchell “model details”; EU Annex IV 1(a), Annex XI 1(c), Article 13(3)(a); SB 53 22757.12(c)(1)(A) to (C); RAISE 1421(3)(a)(i) to (iii); Illinois 10(c)(1)(A) to (C))
  2. Architecture, size, and modalities, with input and output formats, maximum sizes, and languages. (EU Annex XI 1(d) and (e), Annex XII 1(f), 1(g) and 2(b); SB 53 (c)(1)(D) and (E); RAISE 1421(3)(a)(iv) and (v); Illinois 10(c)(1)(D) and (E))
  3. Intended uses and users. (EU Article 13(3)(b)(i), Annex XI 1(a); SB 53 (c)(1)(F); RAISE 1421(3)(a)(vi); Illinois 10(c)(1)(F); Colorado 6-1-1702(1); FDA Appendix E)
  4. Out-of-scope uses, restrictions, and acceptable use policy. (EU Annex XI 1(b), Annex XII 1(b); SB 53 (c)(1)(G); RAISE 1421(3)(a)(vii); Illinois 10(c)(1)(G); Colorado 6-1-1702(1))
  5. Distribution and licence. (EU Annex IV 1(d), Annex XI 1(c) and (f), Annex XII 1(c) and (h))
  6. Integration requirements: hardware, software versions, technical means. (EU Annex IV 1(b), (c) and (e), Annex XII 1(d), 1(e) and 2(a), Article 13(3)(e))
  7. Training data: sources, provenance, selection, size, scope, curation, bias detection. (EU Annex IV 2(d), Annex XI 2(c), Annex XII 2(c); AB 2013 Civil Code 3111(a); Colorado 6-1-1702(1))
  8. Design and training process, with key design choices and rationale. (EU Annex IV 2(a) and (b), Annex XI 2(b))
  9. Training compute and energy. (EU Annex XI 2(d) and (e))
  10. Evaluation: metrics, test data, results overall and by group. (Mitchell “quantitative analyses”; EU Article 13(3)(b)(ii) and (v), Annex IV 2(g), 3 and 4; FDA Appendix E; NIST MEASURE 2.9)
  11. Known limitations, risks, and foreseeable misuse. (EU Article 13(3)(b)(iii), Annex IV 3; Colorado 6-1-1702(1); FDA Appendix E; NIST MAP 2.2)
  12. Human oversight and interpretation of output. (EU Article 13(3)(b)(vii) and (d), Annex IV 2(e); Colorado 6-1-1702(1); NIST MAP 2.2)
  13. Catastrophic risk summaries for large frontier developers. (SB 53 (c)(2); RAISE 1421(3)(b); Illinois 10(c)(2) and (3))
  14. Change log: versions, pre-determined changes, update notices, lifetime, maintenance. (EU Article 13(3)(c) and (e), Annex IV 2(f) and 6; Colorado 6-1-1702(2) and (4); Code Measure 1.1)
  15. Redaction notes and request contact. (SB 53 22757.12(f); RAISE 1421(5); Illinois 10(g); Code Measure 1.2)

How do you build a model card program?

  • List every model you train, fine-tune, or place on the market, and record which rows of the table reach each one
  • Plan for the three audiences these laws name: the regulator file, the deployer or downstream package, and the public report or card
  • For EU high-risk systems, from December 2, 2027 (Annex III) or August 2, 2028 (Annex I Section A), draw up Annex IV technical documentation before placing the system on the market or putting it into service, keep it up to date and at the disposal of national competent authorities for 10 years, and give deployers Article 13 instructions for use
  • For general-purpose models, draw up and keep up to date the Annex XI and XII content, and make the Annex XII information available to providers that intend to integrate the model, unless the open-source exception applies (it never covers models with systemic risk), and, if you sign the Code of Practice, publish a contact for documentation requests
  • Publish any SB 53 transparency report before, or concurrently with, deploying a new or substantially modified frontier model, and calendar the RAISE Act and Illinois’s AI Safety Measures Act for January 1, 2027
  • For SB 53, RAISE, and Illinois documents, describe the character and justification of every redaction to the extent the reason for redacting permits, and keep the unredacted information for five years
  • If you develop covered automated decision-making technology, build the Colorado 6-1-1702 deployer package, a material-update notice process, and three-year record retention before January 1, 2027
  • Put source-attribute-level documentation into health IT contracts before the certification requirement ends December 28, 2026

The AI impact assessment guide covers the deployer assessment a model card feeds, and the EU high-risk AI systems guide covers the provider duties behind Annex IV.

This page is general information as of November 3, 2026, not legal advice; confirm current text with counsel or the regulator before relying on it.

Common questions

Is a model card legally required?

No statute covered in this guide requires a document called a model card. California SB 53 and, from January 1, 2027, New York's RAISE Act and Illinois's AI Safety Measures Act require frontier developers to publish a transparency report before, or concurrently with, deploying a new or substantially modified frontier model, and a developer that publishes that information as part of a larger document, including a system card or model card, is deemed in compliance with the applicable paragraph. The EU AI Act requires model documentation under other names: Annex IV technical documentation and Article 13 instructions for use for high-risk systems, and Annex XI and XII documentation for general-purpose AI models.

What must a general-purpose AI model provider document under the EU AI Act?

Since August 2, 2025, Article 53(1) has required technical documentation of the model, including its training and testing process and the results of its evaluation, with at least the Annex XI information, to provide on request to the AI Office and national competent authorities, and information and documentation with at least the Annex XII elements for providers of AI systems that intend to integrate the model. Providers of models released under a free and open-source licence that allows for the access, usage, modification, and distribution of the model, and whose parameters, including the weights, the information on the model architecture, and the information on model usage, are made publicly available, are exempt from those two duties unless the model has systemic risk. Models placed on the market before August 2, 2025 have until August 2, 2027.

Does the ONC health IT model card requirement still apply?

Only until December 28, 2026. The HTI-5 final rule, published October 28, 2026 at 91 FR 67916, removes the decision support intervention source attribute, source attribute access and modification, and intervention risk management requirements in 45 CFR 170.315(b)(11)(iv) to (vi) effective that date. Separately, FDA's January 2025 draft guidance on AI-enabled device software functions includes an example model card in Appendix E and says FDA does not require a model card or a specific model card format.

By industry: all 8 industry pages

Sources

  1. Model Cards for Model Reporting (arXiv 1810.03993), FAT* ’19, arXiv, submitted October 5, 2018, revised January 14, 2019.
  2. Regulation (EU) 2024/1689 (Artificial Intelligence Act), consolidated text as of July 27, 2026, as amended by Regulation (EU) 2026/1744, EUR-Lex, Articles 2, 11, 13, 18, 53, 111, and 113 and Annexes IV, XI, and XII, accessed November 3, 2026.
  3. Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex, July 24, 2026.
  4. The General-Purpose AI Code of Practice, European Commission, accessed November 3, 2026.
  5. Code of Practice for General-Purpose AI Models, Transparency Chapter (PDF), European Commission, July 10, 2025.
  6. Model Documentation Form (DOCX), European Commission, July 10, 2025.
  7. SB-53 Artificial intelligence models: large developers (Chapter 138, Statutes of 2025), California Legislative Information, 2025.
  8. AB-2013 Generative artificial intelligence: training data transparency (Chapter 817, Statutes of 2024), California Legislative Information, 2024.
  9. California Legislative Information bulk data, 2025-26 session (pubinfo_2025.zip): current text of Business and Professions Code 22757.11 and 22757.12 and Civil Code 3110 and 3111, California Legislative Counsel, extract dated November 2, 2026.
  10. Court calendar: X.AI LLC v. Bonta, No. 26-1591, November 18, 2026, U.S. Court of Appeals for the Ninth Circuit, accessed November 3, 2026.
  11. S6953B (RAISE Act), bill status and text, signed as Chapter 699 of 2025, New York State Senate, accessed November 3, 2026.
  12. A9449 (chapter amendment to the RAISE Act), signed as Chapter 96 of 2026, New York State Senate, accessed November 3, 2026.
  13. Public Act 104-0538 (SB 315), Artificial Intelligence Safety Measures Act (PDF), Illinois General Assembly, approved July 6, 2026, effective January 1, 2027, Sections 5, 10, 18, 25, and 99.
  14. Health Data, Technology, and Interoperability: ONC Deregulatory Actions To Unleash Prosperity (HTI-5 final rule), 91 FR 67916, Federal Register, October 28, 2026.
  15. 45 CFR 170.315, ONC certification criteria for health IT, eCFR, up to date as of October 30, 2026.
  16. Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (draft guidance), U.S. Food and Drug Administration, January 7, 2025.
  17. Artificial Intelligence-Enabled Device Software Functions, draft guidance (PDF), Appendices E and F, U.S. Food and Drug Administration, January 2025.
  18. SB 26-189, Session Laws 2026, Chapter 131 (enrolled act), Colorado General Assembly, approved May 14, 2026.
  19. X.AI LLC v. Weiser, No. 1:26-cv-01515 (D. Colo.), docket, minute order ECF No. 24 (April 27, 2026), CourtListener RECAP, accessed November 3, 2026.
  20. Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (PDF), National Institute of Standards and Technology, January 26, 2023.
  21. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1 (PDF), National Institute of Standards and Technology, July 26, 2024.
  22. ISO/IEC 42001:2023, AI management systems, International Organization for Standardization, December 2023.