Two announcements this week put the same idea into practice at very different scales: when a serious AI incident happens, someone outside the developer should hear about it quickly.
New York: register, then report
A release from the New York State Department of Financial Services, dated November 11, 2026, says the new Office of Digital Innovation, Governance, Integrity, and Trust (DIGIT) "launched its online portal for large frontier AI developers to register with New York State by January 1, 2027." It adds that, beginning in January, large frontier developers "will also be required to publish safety frameworks on their websites, and all frontier developers will be required to report critical safety incidents and publish model transparency reports."
- Who is large. Under the 2026 chapter amendment, Senate Bill S8828, a large frontier developer is one that, with its affiliates, had more than $500 million in annual gross revenue in the preceding calendar year, according to Morrison Foerster's analysis.
- How fast. The same analysis says critical safety incidents must be reported within 72 hours, or within 24 hours to an appropriate authority if an incident poses an imminent risk of death or serious physical injury.
- Who checks. The release says a new state research centre at Binghamton University, NYCAIRR, will coordinate with DIGIT "to begin reviewing AI Developers' published safety frameworks".
Washington and Beijing: an alert line
In remarks to the National Committee on U.S.-China Relations on November 10, Bessent said the US has begun "a new dialogue with China on SI", his abbreviation for superintelligence, "designed to guard against shared risks and prevent a misunderstanding between our two systems." He said both sides expect to "exchange best practices for evaluating and testing advanced models before their release", and that "we are establishing a direct channel to ensure that either nation can promptly alert the other to serious incidents involving its SI models." He also said the two countries should "harbor no illusions about the differences" between them. CNBC notes that neither government has committed to limiting how fast AI advances.
What remains uncertain. New York has not said how it will review registrations or define a reportable incident in practice beyond the statute. The US-China channel has no published scope, operators or thresholds, and an alert line is only as good as the detection behind it.
Analysis: Most teams that build on frontier models are nowhere near the $500 million threshold, but the duties flow downstream: providers will ask customers for faster incident information. Belgavi's Book Store does not train frontier models, yet Geetha could still adopt the same habit at small scale, with a written definition of a serious agent incident, such as a refund issued against the wrong order, a named person who must hear about it within hours, and a log kept well enough to answer a provider's questions.
Sources and evidence
- Governor Hochul Declares New York Will Lead the Nation in AI Science and Safety; Launches New Research Institute (NY Department of Financial Services via EIN Presswire, November 11)
- New York Amends the RAISE Act to Align More Closely with California Law (Morrison Foerster, April 3)
- Remarks from Secretary of the Treasury Scott Bessent at the National Committee on U.S.-China Relations Gala (U.S. Treasury, November 10)
- U.S., China should 'harbor no illusions' as they cooperate, compete on AI, Bessent says (CNBC, November 11)
New York's DIGIT office opened a registry for large frontier AI developers, who must sign up by January 1, 2027, ahead of 72-hour incident reporting and published safety frameworks under the RAISE Act. Bessent said the US and China are building a channel to alert each other to serious incidents with advanced models. Neither limits development. Define and route your own incidents now.