New York Attorney General Letitia James announced on November 17 the creation of the AI Safety, Standards, Enforcement, and Transparency Unit, known as ASSET. The unit sits inside the Office of the Attorney General's Economic Justice Division, works with the Bureau of Internet and Technology, and is led by Cristy Phillips, formerly Special Counsel for the Economic Justice Division.

The timing is the story. The unit's first major statute, the RAISE Act, does not take effect until January 1, 2027. New York is building the enforcement capacity before the law starts, which is unusual in a field where regulators often spend the first year after a statute figuring out what it means.

The laws on its plate

The attorney general's release lists a portfolio that reaches across several kinds of AI harm. The RAISE Act, the Responsible AI Safety and Education Act, requires large frontier-model developers to make public disclosures about their safety measures and to report serious incidents. It takes effect January 1, 2027.

The chatbot laws come next. S.9051B/A.10379C prohibits unsafe chatbot features for minors and takes effect in June 2027. General Business Law Article 47 requires chatbots to identify and respond to suicide risk and to tell users they are not speaking with a person. The SHIELD Act requires safeguards against data breaches. Penal Law Section 245.15 addresses the sharing of intimate images to cause harm, including AI-generated images. The Synthetic Performer Disclosure Act requires advertisements to disclose when AI is used in place of a person.

That is a broad mandate for a unit that is still being staffed. The release does not announce fine amounts or say how many lawyers and technologists the unit will have. It does say the unit will develop policy proposals and provide technical expertise to investigations across the office.

Why an enforcement unit

A law on paper needs an institution to apply it. The ASSET Unit is designed to be that institution for AI in New York. The unit's placement in the Economic Justice Division signals that the attorney general sees AI harm as a consumer and market problem, not only a technology problem. The connection to the Bureau of Internet and Technology gives it access to the office's existing privacy and online-platform work.

The office has been building toward this for months, Bloomberg Law reported. It launched a whistleblower portal for technology workers who want to report unsafe AI development, and it has issued consumer alerts about AI-driven scams. The unit consolidates those threads and gives them a permanent home.

The whistleblower channel is worth watching. Employees inside AI companies see design decisions and incident responses that never become public. A portal that promises confidentiality gives the attorney general a source of information that is difficult to obtain through a document request. The unit's cases may begin with a complaint rather than with a market study.

The backdrop is a federal government that has not settled on a comprehensive AI enforcement regime, as DataGuidance noted in its coverage of the unit. State attorneys general have been filling the gap, using consumer protection, privacy, and anti-discrimination statutes. New York's unit is the most explicit version of that strategy so far: a dedicated team with a named portfolio and a leader.

That strategy has a logic. A state attorney general can investigate a company, issue subpoenas, negotiate a settlement, and seek civil penalties without waiting for a federal agency to act. The statutes are often broad, and the standards are often defined through enforcement rather than rulemaking. For a technology that is changing faster than the federal regulatory process, that speed is the point.

The strategy also has limits. A state attorney general's jurisdiction stops at the state line, and a national company may treat a single state's action as a cost of doing business. The multistate coalition is the way around that limit. If several attorneys general adopt the same theory, the cost rises. New York's unit is likely to be a source of theories that other states borrow.

How the unit will work

Attorney general enforcement usually starts with an investigation. The office can request documents, interview witnesses, and issue subpoenas. If it finds a violation, it can negotiate a settlement, seek an assurance of discontinuance, or file a case. The remedies can include civil penalties, injunctive relief, and changes to business practices.

For AI cases, the evidence will be different from a traditional consumer fraud case. The central questions will be about design choices, disclosures, and incident response, not about a single misleading advertisement. That is why the unit's technical expertise matters. A lawyer who cannot read a model card or a system prompt will struggle to prove that a company failed to disclose something a regulator thinks it should have disclosed.

The staffing question is open. The office has posted jobs for the unit, and the release says it is still being built. A small team can set priorities and bring cases. A larger team can run parallel investigations and produce guidance. The first year will show which model the attorney general chose.

The RAISE Act

The RAISE Act is the unit's anchor. It applies to large frontier-model developers, the companies building the most capable general-purpose systems. The law requires public safety disclosures and incident reporting. The details of the statute, including the thresholds that define a covered developer and the content of the safety plan, will determine how much work the unit has.

The release's description is general, and the enforcement questions are specific. What counts as a serious incident? How quickly must it be reported? What has to be in the safety plan, and who audits it? The unit will have to answer those questions through guidance and, eventually, through cases. The first enforcement action will matter more than the announcement because it will show how the office reads the statute.

The penalties are also an open question. The release does not list fine amounts. In practice, the deterrent effect of an AI law depends on the size of the penalty relative to the company's revenue, and on whether the attorney general can move quickly enough to matter.

The chatbot provisions

The chatbot laws raise design questions that are harder than they look. Article 47 requires a chatbot to identify and respond to suicide risk and to tell users they are not speaking with a person. The first requirement is a safety feature. The second is a disclosure. Both depend on how the system is built and how it is presented.

The minors' law, effective in June 2027, prohibits unsafe chatbot features for children. It will require age assurance, content controls, and a design review that many products have not done. The attorney general's unit will be enforcing a law that touches product design, not just marketing claims.

The Synthetic Performer Disclosure Act adds another layer. An advertisement that uses AI to stand in for a person has to say so. The unit will have to decide what counts as a disclosure, where it has to appear, and how it applies to the many forms of synthetic media that advertising now uses.

The patchwork

New York is not the only state moving on AI. California, Colorado, and other states have passed or proposed AI laws, and the federal government has its own evolving approach. A company that operates nationally faces a patchwork of definitions, deadlines, and enforcement styles.

That patchwork is the reason a dedicated unit matters for companies, not just for New Yorkers. An enforcement action in one state can set a template that other attorneys general follow. The ASSET Unit's public guidance and its first cases will be read outside New York, and the office knows it.

The unit also gives companies a counterpart to talk to. A firm that wants to know how the attorney general reads the RAISE Act can ask the unit. That is not the same as legal certainty, but it is more than the silence that often follows a new statute.

What companies should do

Start with an inventory. Which products use frontier models, chatbots, synthetic media, or AI-generated advertising? Which of those products touch minors, health, finance, or personal data? The answers determine which of the unit's statutes apply.

Then check the documentation. The RAISE Act's safety-disclosure and incident-reporting duties require records that many companies do not keep in a form a regulator can read. An incident-response plan that covers AI failures is part of the same work. So is a vendor review that asks what the model provider will disclose and when.

The vendor review is the part companies most often skip. A developer that builds on a frontier model may not control the model's safety testing, but it does control what it asks for, what it documents, and what it tells users. A contract that requires the model provider to share incident information and safety documentation is a reasonable response to a law that puts the disclosure duty on the developer.

Then check the consumer-facing pieces. A chatbot that does not identify itself, or does not handle a suicide-risk disclosure, is exposed under Article 47. An ad that uses a synthetic performer without a disclosure is exposed under the Synthetic Performer Disclosure Act. A product that is unsafe for minors is exposed under the June 2027 law.

The first case

The ASSET Unit will be judged by what it does, not by what it announces. A unit with a broad mandate and a small staff can produce guidance and referrals. A unit with a real enforcement record can change how companies design products.

The attorney general's statement framed the mission around safety and privacy. The statutes give the unit the tools to pursue both. The first case will show which tool it reaches for, and that choice will tell companies more than the press release does.

Until then, the unit's most useful output may be guidance. A set of published expectations about safety disclosures, incident reports, and chatbot behavior would let companies comply before an investigation begins. The office has not promised that guidance, and the absence of it is itself a signal about how the first year will go.

Primary sources

  1. Office of the New York State Attorney General, press release of November 17, 2026, for the unit's name, leadership, placement, and the list of laws.
  2. DataGuidance, for the coverage of the ASSET Unit and the RAISE Act's effective date.
  3. Bloomberg Law, for the enforcement scope and the unit's role in the attorney general's office.