Two recent developments show that New York and California are making different choices about who should have to disclose the use of AI. On September 30, 2026, California Governor Gavin Newsom signed SB 574, which requires lawyers to tell courts when they used generative AI to prepare a filing. New York decided earlier this year not to impose that duty on lawyers. At the same time, a bill that has passed the Senate and the Assembly (A6578B) would require AI developers to publish information about the data used to train their models, following a law California adopted in 2024.
The pattern is worth noticing. New York doesn’t ask lawyers to disclose how they drafted a brief, because they already answer for every word they file. It may soon ask the companies building AI tools to disclose what went into them.
Lawyers and Court Filings
California’s SB 574 appears to be the first state statute on how lawyers use generative AI. It will take effect on January 1, 2027. Lawyers may not delegate the practice of law to AI, must personally verify every citation before filing, and may not put confidential information into public AI tools. They must also disclose the use of generative artificial intelligence to the court for all documents submitted to the court.
New York went the other way. Since June 1, 2026, Part 161 of the court system’s rules (22 NYCRR Part 161) has allowed lawyers to use AI in preparing court papers and states that they should not have to disclose it. Since June 1, 2026, Part 161 of the court system’s rules (22 NYCRR Part 161) has allowed lawyers to use AI in preparing court papers and states that they should not have to disclose it. Individual courts may adopt their own AI rules, so the assigned judge’s part rules matter.
Where the two states agree is verification, and for good reason. A database kept by researcher Damien Charlotin counts more than 1,500 U.S. decisions involving AI-fabricated material in filings, most of them by self-represented litigants. Sanctions have grown from $5,000 in Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), to monetary sanctions plus a fee award of more than $94,000 in an Oregon federal case (Couvrette v. Wisnovsky, No. 1:21-cv-00157-CL, 2025 U.S. Dist. LEXIS 273533 (D. Or. Dec. 12, 2025).
Why New York Said No to Lawyer Disclosure
The main reason is that the duties already exist: Part 161 says the obligations owed to a court apply to every submission, whether or not AI was used.
Part 161 relies on review. Its model rule, which individual courts may adopt, requires a lawyer who uses AI to carefully review the paper and independently ensure it contains no fabricated or fictitious cases, statutes, or other material, and treats the signature as certifying that review. Even where a court has not adopted it, the same obligation follows from Part 130 and the Rules of Professional Conduct. Courts have used these rules to punish AI misuse, although sanctions cannot prevent the tools from fabricating authority. In January 2026, the Third Department sanctioned counsel under Part 130 in Deutsche Bank National Trust Co. v. LeTennier, the first New York appellate decision to do so for AI-fabricated cases.
There is also a practical problem. AI is now built into ordinary tools like Word, Westlaw, and Lexis, so a broad disclosure rule would be hard to apply. The New York City Bar’s 2024 opinion did not call for disclosure to courts; it told lawyers to review every AI output before filing and to check whether a court’s own rules require disclosure. A court rule is also easier to update than a statute as the technology changes. New York isn't alone. In 2024 the Fifth Circuit considered an AI certification rule and dropped it, reminding lawyers that “I used AI” is no excuse for a sanctionable filing.
AI Developers: New York May Follow California
On the developer side, New York is moving toward disclosure. A6578B, the Artificial Intelligence Training Data Transparency Act, would add a new Article 44-C to the General Business Law. Developers of generative AI models and services made available to New Yorkers would have to post on their websites a high-level summary of the data used to train them. The summary would cover at least twelve points, including where the datasets came from, whether they contain copyrighted or personal information, whether they were bought or licensed, and whether synthetic data was used.
The bill closely tracks California’s AB 2013, which took effect on January 1, 2026. Like the California law, it covers models and services released on or after January 1, 2022, and applies to anyone who substantially modifies a model for public use, including through retraining or fine-tuning. The twelve disclosure items are essentially identical, except that New York’s personal-information item incorporates General Business Law § 899-aaa rather than the CCPA.
The exemptions are close but not identical. Both laws exclude models whose sole purpose is operating aircraft in the national airspace and defense models made available only to a federal entity. New York, however, leaves out California’s exemption for models whose sole purpose is security and integrity.
Timing is less certain. The bill received final legislative passage on June 4, 2026, when the Senate approved it 54–6, but it has not yet been delivered to Governor Hochul, and her decision period runs from delivery. If signed as written, it would take effect immediately, with a first disclosure deadline of January 1, 2027, which leaves developers little time. The governor could also sign it subject to negotiated chapter amendments, as she did with the RAISE Act, or veto it.
For developers that already publish California disclosures, much of the work should carry over, starting with confirming that existing documentation also covers models offered in New York. The personal-information item needs closer attention. Because A6578B incorporates § 899-aaa rather than the CCPA, developers cannot automatically rely on California’s treatment of publicly available information. Developers that concluded their datasets contained no CCPA-defined personal information solely because the information was publicly available should reassess that conclusion under New York’s definitions. The relatively few developers that relied on California’s security exemption would need to prepare disclosures from scratch.
The California law is also being tested in court. xAI sued to block AB 2013, arguing, among other things, that it forces disclosure of trade secrets and compels speech. A federal district court denied a preliminary injunction on March 4, 2026, and xAI’s appeal is pending before the Ninth Circuit, with oral argument calendared for November 18, 2026. A Ninth Circuit ruling would not bind New York courts, but as the first appellate decision on these challenges, it would carry significant persuasive weight if New York enacts its version.
What This Means
For lawyers, the rule is simple in both states. Verify every case, statute, and quotation before you sign. Starting in 2027, California filings prepared with generative AI will also need a disclosure. New York state court filings generally won’t, unless the assigned judge requires it, although bills pending in Albany (A.8546/S.9794 and S.2698) would require a disclosure affidavit in civil filings. Courts have also been far harder on lawyers who deny using AI or blame someone else than on those who admit a mistake.
For companies that build or fine-tune AI models offered to the public, the question is whether Governor Hochul signs A6578B. If she does, the training data disclosures many developers already post for California will become a New York requirement as well.