A government portal registration screen.
By Ellis Ward | November 11, 2026
The Law Is Live. Enforcement Is Not.
New York's RAISE Act registration portal opened November 10. The law's enforcement period does not begin until January 1, 2027. That 52-day gap between portal opening and first enforcement date is not a minor procedural detail. It is, in practical terms, the entire current state of the law: companies can register now, they must register by January 1, and they face no penalty under the new regime until that date arrives.
That distinction matters for anyone trying to understand what New York's AI safety law actually requires today, and what it does not yet require. The RAISE Act was signed December 19, 2025. An amending bill, S8828, revised definitions and penalties, passed the Assembly on March 11, 2026, and was signed on March 27. That amendment moved the effective date to January 1, 2027. The result is a law that is fully enacted and partially operational, but not yet enforcing.
The Office of Digital Innovation, Governance, Integrity, and Trust, known as DIGIT, will implement and oversee the law. DIGIT sits within New York State's Department of Financial Services. As of Tuesday, DIGIT had not publicly identified which companies it considers covered registrants. It had not publicly stated how it will interpret the incident-reporting clock. No company has yet faced a penalty under the new regime.
Who the Law Actually Covers
The RAISE Act creates two overlapping categories of obligation. Some requirements apply to all frontier model developers, regardless of company size. Others apply specifically to what the law calls "large frontier developers."
The frontier-model threshold turns on training compute: more than 10^26 operations. Any company that trains a model at that level falls under the universal requirements. Those requirements are three: publish safety frameworks on the company's website, report critical safety incidents within 72 hours of occurrence, and issue public model transparency reports. These rules apply whether or not the company generates $500 million in annual revenue. The law does not make them optional for smaller developers.
Large frontier developers carry the additional obligation to register with the state. The revenue threshold is annual revenue above $500 million. For those companies, registration is mandatory. The deadline to register is January 1, 2027. The portal to do so opened November 10, giving large frontier developers 52 days to file before the enforcement window opens.
Two categories are explicitly excluded. Academic research at accredited in-state universities is exempt. The Empire AI Consortium, a $500 million public-private computing initiative, is also exempt. Those carve-outs reflect a deliberate choice to focus the law on commercial frontier model development while protecting the state's own research infrastructure from the compliance burden it has asked of private companies.
What DIGIT Will Do, and What Remains Undefined
DIGIT will serve as the law's enforcement arm. The state can bring enforcement actions for failure to register. Penalties run $1 million for a first violation and $3 million for each subsequent violation. Those figures were set by S8828 before it was signed in March.
What DIGIT has not yet publicly established is substantial. The office has not named which specific companies it will designate as covered registrants. It has not publicly explained what verification process will precede enforcement. It has not publicly stated how it will interpret the 72-hour incident-reporting clock: when that clock starts, what constitutes a reportable critical safety incident, and what documentation companies must provide. These are not minor procedural gaps. They determine whether the law functions as written or settles into a looser set of expectations.
Governor Hochul's administration also launched the New York Center for AI Responsibility and Research at SUNY Binghamton University. The center will independently evaluate the safety, security, and operation of advanced AI systems. It will work alongside DIGIT to review companies' safety frameworks. The co-directors, announced on an interim basis, are Greg Morrisett of Cornell Tech and Jeannette Wing of Columbia University. The center plans to bring on more than 80 researchers and will draw computing resources from the Empire AI initiative.
The center has no legal authority to require company audits. Its role is evaluative. It can review safety frameworks and produce findings, but it cannot compel a company to submit to an audit. Hochul announced the center in January with $55 million in donations and capital.
At Tuesday's press briefing, Hochul stated her framing directly. "Tech moguls tell us that their vision for AI is inevitable," she said. That line sets the tone for the administration's posture. The law positions New York as a check on that framing. How firm a check depends on enforcement decisions DIGIT has not yet made public.
What Companies Have Said, and What They Haven't
Neither OpenAI nor Anthropic had publicly responded to the registration opening as of Tuesday. That silence is notable. Both companies build frontier models. Both fall within the categories the RAISE Act addresses. The January 1 deadline is now less than two months away.
Anthropic's policy page states that the company "has advocated for safety-transparency laws, including RAISE." That language is Anthropic's own. It reflects the company's position on a category of law. It does not say Anthropic endorses every provision in the RAISE Act's current form, and it should not be read that way. Advocacy for a type of law and acceptance of a specific law's terms are not the same thing.
OpenAI had not published a public position on the RAISE Act as of Tuesday. No other large frontier developer's position was available from the verified sources used for this article.
New York is not the only state writing at this scale. California's SB 53 took effect January 1, 2026. Illinois's SB 315 takes effect January 1, 2027. Illinois matches New York on the revenue threshold, the 72-hour incident-reporting deadline, and the penalty cap. That convergence suggests either coordination among states or shared modeling from common policy sources, or both. It also means that large frontier developers now face nearly identical obligations in two states simultaneously.
Two AI Laws in One Year, and a Different Debate in Dutchess County
The RAISE Act is not the only AI law Hochul signed this year. In October, she signed a separate chatbot safety law for minors. That law restricts deceptive human-like interactions, emotionally manipulative or sycophantic responses, and encouragement of suicide or disordered eating. It takes effect June 1, 2027. The two laws address different problems. The RAISE Act targets frontier model development at the company level. The chatbot law targets consumer-facing systems deployed to younger users.
For Dutchess County, the RAISE Act arrives against a backdrop of ongoing local debate about AI and infrastructure. According to RFTUV's prior reporting, the Dutchess County AI Oversight Committee last met on June 4, 2026. Those minutes covered county AI policy and a proposed transparency bill for AI data-center and high-consumption projects. No November meeting record had surfaced as of the time this article was submitted.
The committee's data-center transparency work represents a parallel track. It is local oversight of AI's physical infrastructure, running alongside Albany's attempt to regulate the companies that build the models. The June minutes predate the RAISE Act's registration portal by five months. What the committee's members make of the new state framework, once the committee reconvenes, is an open question.
The gap between local and state action also points to a gap in scope. Dutchess County's proposed transparency bill addressed high-consumption AI projects at the county level. The RAISE Act operates at the frontier-developer level, statewide. Whether those two frameworks interact, conflict, or simply address separate problems without touching each other has not been publicly addressed by either the county committee or the state.
What the 52-Day Window Actually Means
Starting November 10, large frontier developers can register with the state. Starting January 1, they must have registered, or face a $1 million penalty for the first failure. In the 52 days between those two dates, the law is real but its enforcement provisions are not yet active.
What companies do in this window is worth watching. Registration is not simply an administrative form. It places a company on record as a covered entity. The public requirements that follow, safety frameworks on company websites, 72-hour incident reporting, transparency reports, are attached to that covered status. Those obligations activate at enforcement, not at registration, but registration is what starts the relationship with DIGIT and, eventually, with the SUNY Binghamton research center.
The undefined elements will shape how much any of this matters in practice. How DIGIT interprets the incident-reporting clock determines whether the 72-hour rule is a live obligation with real consequences or a vague standard companies can satisfy on their own terms. Which companies DIGIT names as covered registrants determines whether the law reaches the developers actually building the most capable systems. What verification looks like before enforcement determines whether January 1 is a firm deadline or a soft one.
None of those questions have been answered publicly. New York opened the portal. It has not yet explained, in detail, what it will do with what comes through it.
Sources
- https://techcrunch.com/2026/11/10/openai-anthropic-face-new-oversight-as-new-yorks-ai-safety-law-takes-effect/
- https://xenospectrum.com/en/raise-act-registration-research-center/
- https://dailyinference.news/articles/2026-11-10-skip-an-ai-safety-filing-in-new-york-and-the-fines-start-at-1-million/
- https://www.ainewstoday.net/new-york-can-fine-openai-and-anthropic-up-to-3-million-dollars-under-new-ai-law
