New York Begins Safety Checks for Large AI Models
- Governance e compliance

The State of New York has opened the portal through which major developers of advanced models will have to register by January 1, 2027, effectively launching enforcement of the RAISE Act, one of the most significant state laws in the United States on artificial intelligence safety and transparency. The measure applies to developers with more than $500 million in annual revenue and introduces specific obligations: publishing a framework of safety practices on their website, making public model transparency reports available, and notifying critical incidents within 72 hours.
Implementation will be entrusted to the new Office of Digital Innovation, Governance, Integrity, and Trust, which will have oversight functions over major frontier AI labs. The law provides for penalties of up to $1 million for a first violation and up to $3 million for subsequent ones. Alongside the state office, the New York Center for AI Responsibility and Research is also being established, with the task of independently assessing the safety, protection, and operation of advanced systems and collaborating in the review of the frameworks published by companies.
For enterprise audiences, this news matters because it marks a shift from regulatory discussion to the operational rollout of concrete obligations. The issue is no longer only general principles on trustworthy AI, but verifiable processes: registration, documentation of safety measures, formal incident management, and public accountability for the models released. Even for organizations that do not develop foundation models, this kind of rule raises the bar on governance expectations across the supply chain: vendor selection, contractual clauses, escalation procedures, and decision traceability are becoming increasingly difficult to handle informally.
The significance of the news lies not only in its impact on OpenAI or Anthropic, mentioned among the affected parties, but in the precedent it creates. TechCrunch notes that other states, such as California and Illinois, have already introduced or are introducing similar requirements, albeit with differences in timing and enforcement mechanisms. For those operating in regulated sectors, the key point to watch is the convergence toward supervision models that require documentary evidence and defined response times in the event of an incident.
In this context, the ability to govern AI systems with clear roles, process controls, and structured exception handling becomes an operational requirement, not just a legal one. It is the kind of scenario in which agentic architectures with human oversight and native traceability can help make AI adoption more orderly in highly compliant environments.
Source: TechCrunch AI, 10 November 2026