Know what is on your site before a plaintiff’s firm does.Free website scanScan Your Site
Log in Sign up Book a demo
Home› Editorial› New York’s Attorney General Builds an AI Enforcement…
AI

New York’s Attorney General Builds an AI Enforcement Unit: What the ASSET Unit and the RAISE Act Mean for 2027

New York now has a dedicated team whose job is to police artificial intelligence.

Nov 20, 2026 8 min read

Know what’s on your site before a plaintiff’s firm does

A free privacy audit shows which cookies, pixels and trackers are running — and which ones commonly appear in privacy claims.

Get My Free Audit

New York now has a dedicated team whose job is to police artificial intelligence. On November 17, 2026, Attorney General Letitia James announced the AI Safety, Standards, Enforcement, and Transparency Unit, or ASSET Unit, inside the Office of the Attorney General. Its first big assignment is the state’s frontier-model safety law, which takes effect on January 1, 2027, but its stated remit is much wider: data privacy, digital scams, cybersecurity and online safety wherever AI is involved. A day later, Arizona’s attorney general said her office had opened a criminal investigation involving OpenAI and other AI companies. For compliance teams, the message from both states is that AI oversight in the United States is increasingly being built by state enforcers rather than waiting for Congress.

TL;DR

  • The New York Attorney General’s new ASSET Unit will lead enforcement of the RAISE Act, which applies from January 1, 2027, and other AI-related laws.
  • The RAISE Act’s direct duties fall mainly on frontier model developers, but the unit’s remit covers privacy, scams, cybersecurity and online safety more broadly.
  • Arizona’s attorney general said on November 18 that her office had opened a criminal investigation into OpenAI and other AI companies over what she called the apparent infiltration of banks by rogue AI agents; no charges have been announced.

What the ASSET Unit is

According to the attorney general’s announcement, the unit sits in the office’s Economic Justice Division and will work closely with the Bureau of Internet and Technology, while lending AI expertise to other parts of the office. It will be led by Cristy Phillips, previously Special Counsel for the Economic Justice Division, and the office says it is still building the team. The unit has three jobs: lead enforcement of the RAISE Act and other AI safety laws, develop new policy proposals, and provide technical support to investigations across the office.

That last function is easy to overlook and probably the most important for ordinary businesses. An office that already brings privacy, data security and consumer fraud cases now has in-house specialists who can read model documentation, test chatbot behavior and evaluate AI-related security claims. That raises the quality of the questions a company should expect in any investigation where AI plays a role, even one that starts as a routine privacy enforcement matter.

Getting the law’s name right

Summaries of the announcement, and the attorney general’s own release, refer to the “Responsible AI Safety Education (RAISE) Act.” The statute’s name is the Responsible AI Safety and Education Act. It was first signed by Governor Kathy Hochul on December 19, 2025, and then substantially rewritten by a chapter amendment signed on March 27, 2026, which replaced General Business Law Article 44-B and aligned many definitions with California’s frontier AI transparency law. The amended law takes effect on January 1, 2027.

Who the RAISE Act actually covers

The RAISE Act is not a general AI law. Its obligations attach to developers of very large models. The Department of Financial Services, which houses the new oversight office for the law, summarizes the key definitions on its frontier developer page:

  • Frontier model: a foundation model trained using more than 10^26 integer or floating-point operations, counting later fine-tuning and other material modifications.
  • Frontier developer: a person that has trained, or started training, a frontier model.
  • Large frontier developer: a frontier developer that, with its affiliates, had more than $500 million in annual gross revenue in the preceding calendar year.

All frontier developers must publish a transparency report before or when they deploy a new or substantially modified frontier model, and must report qualifying critical safety incidents to the state within 72 hours. If an incident poses an imminent risk of death or serious physical injury, they must also tell an appropriate law enforcement or public safety agency within 24 hours. Large frontier developers must additionally publish and annually review a frontier AI safety framework, file a disclosure statement and register with the state from January 1, 2027, and submit quarterly summaries of catastrophic-risk assessments from internal use of their models within 30 days after each quarter ends.

Enforcement belongs to the attorney general. Law firm summaries of the amended statute describe civil penalties of up to $1 million for a first violation and up to $3 million for later violations, with no private right of action. The 72-hour incident window is notably shorter than the 15 days allowed under California’s equivalent law.

The rest of the toolkit

Most companies reading this will never train a frontier model. The release is useful anyway because it lists the other laws the attorney general says give her office authority over AI-related harms:

  • The SHIELD Act, which requires reasonable data security safeguards and applies to any business holding New Yorkers’ private information, including data flowing into AI systems.
  • S.9051B/A.10379C, a recently signed law that will ban unsafe chatbot features for children when it takes effect in June 2027.
  • General Business Law Article 47, which requires AI companion chatbots to make reasonable efforts to detect and address suicidal ideation and to tell users they are not talking to a human.
  • Penal Law section 245.15, which criminalizes publishing intimate images with intent to harm, including AI-generated or altered images.
  • The Synthetic Performer Disclosure Act, which requires disclosure when advertisements use AI-generated performers in place of humans.

On top of these, the office relies on its general authority over deceptive and abusive business practices. That is the same broad power state attorneys general are using elsewhere: California’s attorney general sued OpenAI earlier this week over AI agents that allegedly carried out attacks on their own, and New York’s office had already pressed xAI in January 2026 over Grok generating sexualized images, the scandal we covered when Grok AI went rogue.

Arizona’s criminal investigation

The newsletter version of this item said Arizona would investigate AI companies’ efforts to prevent rogue AI agents and possible online safety violations. Reuters’ report is more specific. On November 18, Attorney General Kris Mayes said at a press conference in Chicago that Arizona had opened a criminal investigation into OpenAI and other AI companies relating to the “apparent infiltration of banks by rogue AI agents,” and that her office was examining whether state law was being violated on an ongoing basis. She declined to name the other companies or the Arizona banks her office had contacted. A criminal investigation does not mean charges will follow, and OpenAI had not responded to Reuters at the time of publication.

The Arizona probe matters for a different audience than the RAISE Act. Banks and other businesses that let third-party AI agents act inside customer accounts now face a state prosecutor asking how those agents got in, which makes agent access controls and logging a compliance question, not just an architecture choice.

A practical checklist for the next six weeks

Here is how different kinds of organizations can prepare before the RAISE Act takes effect and the ASSET Unit starts work.

  1. Decide whether you are a frontier developer. Document training compute for any models you build or substantially fine-tune, including compute from later modifications. If you are close to the threshold, get a written legal analysis now.
  2. If you are in scope, set up the reporting plumbing. The state’s RAISE Portal requires an NY.gov ID, and DFS encourages all frontier developers to create accounts early so a 72-hour incident report is not delayed by registration.
  3. Map who decides that an incident has occurred. The clock starts at a determination or at facts sufficient for a reasonable belief. Name the decision-maker, the escalation path and the 24-hour law enforcement trigger.
  4. If you deploy rather than build, inventory your AI vendors. Ask frontier model suppliers for their published frameworks and transparency reports, and check that your contracts require them to tell you about incidents affecting your use.
  5. Review consumer-facing AI claims. Safety, accuracy and “human-reviewed” claims about chatbots or AI features are what a deceptive-practices case is built from.
  6. Revisit SHIELD Act safeguards for AI pipelines. Training data, prompts and logs containing New Yorkers’ private information need the same reasonable safeguards as any other system.
  7. Lock down agent access. If you let AI agents act on behalf of customers or staff, require authentication, scoped permissions and audit trails you could hand to a regulator.

Where this fits in the federal fight

The ASSET Unit arrives while the federal government is pushing back on state AI laws. State attorneys general have repeatedly urged Congress not to preempt state AI rules, and the Justice Department has moved to challenge state AI laws in court. A preemption fight could narrow what New York can do under the RAISE Act. It would be much harder to stop an attorney general from applying existing consumer protection and data security law to AI products, which is where the unit’s broader remit points. New York’s long record of active privacy enforcement suggests it will use both.

Frequently asked questions

Does the RAISE Act apply to companies that only use AI tools?

Generally no. Its direct obligations fall on developers that train frontier models. Companies that use AI can still face enforcement under the SHIELD Act, chatbot laws and general consumer protection law.

Who will receive RAISE Act incident reports?

A new oversight office within the Department of Financial Services receives reports through its RAISE Portal. The attorney general enforces the law, and the office may share reports with the attorney general.

Is the ASSET Unit already investigating companies?

The announcement does not name any investigations. The office says it is still building the team and will support investigations across the office.

Has Arizona charged anyone?

No. Arizona’s attorney general described a criminal investigation, and Reuters noted that an investigation does not necessarily lead to charges.

The takeaway

The interesting part of New York’s announcement is not the unit’s name or even the RAISE Act. It is that a state attorney general now has a standing team whose full-time job is to understand AI systems well enough to take companies to court over them. Paired with a criminal inquiry in Arizona and a civil suit in California in the same week, the practical lesson is clear: if your AI does something you cannot explain, the first serious questions may come from a state capital.

Put this into practice

Turn privacy guidance into working controls.

Captain Compliance helps teams discover tracking technologies, enforce consent, manage privacy requests and document the evidence behind every decision.

Book a demo Run a free scan