New York State’s most stringent AI oversight regime to date has officially taken effect. Starting January 1, 2027, frontier AI developers—including OpenAI and Anthropic—must register under the RAISE Act, a new law targeting “large frontier labs” with annual revenues above $500 million. Core obligations under the legislation include publishing publicly accessible safety frameworks, reporting critical system failures within 72 hours, and issuing transparency reports about model operations and impacts. Penalties for noncompliance begin at up to $1 million for a first violation, and escalate to $3 million for repeat offenses. New York’s Office of Digital Innovation, Governance, Integrity, and Trust (DIGIT) will enforce the rules and oversee registration of affected labs.
New structures and responsibilities for AI oversight
The law establishes DIGIT as the regulatory body charged with implementing the RAISE Act, ensuring labs with substantial influence make visible safety and risk-mitigation efforts public. As part of the broader framework, Governor Kathy Hochul has also inaugurated a new Center for AI Responsibility and Research at SUNY Binghamton. Already endowed with $55 million, the center—first announced earlier this year—will work alongside DIGIT to independently evaluate the safety, security, and functioning of advanced AI systems.
Related AI legislation and wider implications
RAISE builds on earlier state laws targeting AI, including requirements that AI companion apps disclose bot identity and monitor for signs of self-harm or suicidal ideation. A separate law enacted in October focuses on minors, banning misleading human-like interactions and emotionally manipulative or dangerous content, effective June 1, 2027. Other states are following suit: California’s SB 53—which became law in January, enacting safety frameworks, whistleblower protections, and 15-day incident reporting—along with Illinois’s SB 315, set to take effect in early 2027 with similar demands and independent audits.
The stakes for developers are high. Under RAISE, expected targets are those AI firms producing frontier models with at least $500 million in revenue, but the law also compels visibility across several dimensions: from incident reporting windows to framework transparency to external evaluation. DIGIT has authority to penalize violations heavily and force registration compliance. Even just failing to enroll in the registration portal can risk enforcement actions—even though precise penalties there are unspecified.
What’s new: The RAISE Act’s 72-hour critical incident reporting requirement is among the fastest legally mandated reporting windows for AI labs. Its public safety framework mandate ensures that labs are held accountable not just for performance, but for ethics, transparency, and foreseeable harms. Meanwhile, with the junior AI oversight laws focused on bot identity and minors already law, New York is stacking policy layers designed for angle-by-angle safeguards across different use-cases.
What this means for OpenAI, Anthropic, and similar labs is a coming regulatory surge: documentation, regular disclosure, and faster incident escalations are no longer voluntary best practices but legally required. The law formalizes what many AI ethicists and policy advocates have pushed for. For labs that have been criticized for opacity and risk, it heightens the imperative to clean house when it comes to transparency and risk mitigation.
Looking ahead: As New York becomes the first state to bring full-scale operational oversight to frontier AI labs, its policy design will likely serve as a template for other states or even federal regulation. Companies must begin adapting now—setting up reporting pipelines, safety protocols, and public frameworks—or face steep penalties. Close watchers should track how DIGIT defines “critical safety incidents,” how rigorously audits are applied, and whether enforcement penalties are issued in practice rather than just existing on paper. These will determine whether the RAISE Act transforms AI safety into enforceable accountability—or becomes a regulatory novelty.
