Key Takeaways

  • OpenAI and Anthropic face registration, disclosure, assessment, and incident-reporting requirements beginning January 1, 2027.
  • New York’s DIGIT office can pursue civil penalties reaching $1 million for an initial violation and $3 million for subsequent violations.
  • The law turns practices such as red-teaming, model evaluations, and mitigation documentation into auditable compliance records.

New York has opened the registration process for large frontier-AI developers covered by the Responsible AI Safety and Education Act, putting OpenAI and Anthropic on a short runway toward formal state supervision. Registration is required by January 1, 2027, when the RAISE Act’s central provisions begin to apply.

The law targets developers with more than $500 million in annual revenue that train models using more than 10 to the 26th operations. That combination is intended to focus oversight on the relatively small group building the most computationally intensive systems, rather than sweeping ordinary enterprise AI applications or smaller startups into the same regime. OpenAI’s frontier models and Anthropic’s Claude are named examples of systems likely to fall within its reach.

According to the New York Department of Financial Services, covered developers are required to register with the Office of Digital Innovation, Governance, Integrity, and Trust, known as DIGIT. They also face requirements for written safety frameworks, public transparency reports, quarterly catastrophic-risk assessments, critical-incident reporting within 72 hours, and disclosure statements submitted at least every two years.

That is a significant shift in how frontier-model governance works. Evaluations, red-team exercises, risk thresholds, and mitigation decisions have often been shaped by voluntary commitments or internal policies. Under the RAISE Act, those activities become records that regulators can inspect and compare with a developer’s public claims. A safety framework that looks polished on a website but does not match internal testing could create legal and reputational exposure.

Registration itself is likely the easiest step in the compliance process. The heavier work involves building repeatable processes around model development, deployment approvals, incident classification, documentation, and escalation. OpenAI and Anthropic may need compliance teams to coordinate closely with researchers, security personnel, legal departments, and product leaders. Quarterly assessments also create an ongoing obligation rather than a one-time filing exercise.

The deadlines could prove especially demanding during a serious incident. Seventy-two hours leaves limited room to determine what happened, preserve evidence, evaluate potential harm, brief executives, and prepare an accurate report. What qualifies as a critical safety incident will therefore become an important operational question, not merely a legal definition. Developers may respond by establishing reporting triggers and decision paths before an event occurs.

Enforcement gives those requirements weight. TechCrunch reports that civil penalties can reach $1 million for an initial violation and $3 million for subsequent violations. New York can also pursue action against developers that fail to register, although the state has not detailed how it will approach every potential registration case.

Oversight will not sit with DIGIT alone. Governor Kathy Hochul also launched the New York Center for AI Responsibility and Research at State University of New York’s Binghamton University. Backed by $55 million in donations and capital announced in January, the center will independently evaluate the safety, security, and operation of advanced AI systems. It will also support DIGIT’s review of developer safety frameworks.

The reach could extend beyond OpenAI and Anthropic. Intellisee identifies Google DeepMind’s systems as another example and says the scope could potentially include Meta and xAI, depending on whether particular models and corporate structures satisfy the statutory thresholds.

New York is also building a broader state-level AI rulebook. A companion-technology law already requires operators to disclose that users are interacting with bots and to make reasonable efforts to detect suicidal ideation and self-harm. A separate chatbot law for minors, signed in October and taking effect June 1, 2027, restricts deceptive human-like interactions, emotional manipulation, sycophantic responses, and encouragement of suicide or disordered eating.

The result is a growing compliance patchwork. California’s SB 53 requires published safety frameworks, whistleblower protections, and critical-incident reports within 15 days. Illinois’ SB 315 takes effect at the start of 2027 and adds independent compliance audits. For OpenAI, Anthropic, and other covered developers, the practical response may be a common governance baseline that satisfies the strictest overlapping obligations while preserving state-specific reporting procedures.