New York has begun enforcing its comprehensive AI safety law, compelling leading developers such as OpenAI and Anthropic to complete registration, disclose safety protocols and report critical incidents under threat of financial penalties.

The law, which passed the state legislature last year, aims to create transparency and accountability in the rapidly expanding artificial intelligence sector, and its rollout marks the first major state-level regulatory action of its kind in the United States.

What the New York AI Safety Law Requires

The statute sets out three core obligations for AI developers that meet certain size and impact thresholds. First, companies must file a registration with the New York Department of Consumer and Worker Protection (DCWP) detailing their corporate structure, the AI systems they offer and the intended user base. Second, they must submit a written safety plan that outlines risk-assessment methods, testing procedures, and mitigation strategies for harms such as bias, misinformation and privacy breaches. Third, developers are required to report any incident that the law defines as "critical" - for example, a model output that leads to physical injury, significant financial loss, or widespread disinformation - within 72 hours of discovery.

Failure to comply with any of these obligations can result in civil penalties of up to Rs. 1.6 billion per violation, as stipulated in the legislation. The law also empowers the DCWP to conduct audits and issue cease-and-desist orders if a developer's practices are deemed unsafe.

Background: How New York Arrived at Its AI Safety Framework

New York's regulatory push follows a series of high-profile AI incidents that raised public concern over algorithmic bias, deep-fake misuse and unintentional harms. In 2022, a facial-recognition system used by a city agency misidentified several individuals, prompting lawsuits and calls for stricter oversight. The following year, a large language model generated false medical advice that was shared widely on social media, highlighting gaps in existing consumer protection rules.

State legislators responded by drafting the Artificial Intelligence Accountability Act, which passed the Senate and Assembly with bipartisan support in June 2023. The bill drew on recommendations from the New York Institute for Advanced Technology, a think-tank that had published a 2022 report urging a "risk-based licensing regime" for powerful AI models. After a year of public hearings, stakeholder consultations and amendments, the law was signed by the governor in March 2024 and set to take effect on October 1, 2024.

Explainer: How the Registration and Reporting Process Works

Companies that develop AI systems with more than 10 million active users or that generate outputs with a potential impact on public safety fall under the law's scope. These firms must submit an online registration form through the DCWP portal, providing:

  • Corporate name, headquarters address and ownership details.
  • A catalog of AI models, including version numbers and deployment environments.
  • Projected user demographics and geographic reach.

Within 30 days of registration, developers must upload a safety documentation package. The package should contain:

  1. Risk-assessment methodology, citing specific metrics such as false-positive rates and bias scores.
  2. Testing protocols, including red-team exercises and external audits.
  3. Mitigation strategies, like content filters, human-in-the-loop controls and user-feedback loops.

When a critical incident occurs, the reporting form requires a concise description, the date and time of occurrence, affected users, and steps taken to remediate. The law mandates that the report be filed within 72 hours, after which the DCWP may request additional evidence or launch a formal investigation.

Impact on OpenAI, Anthropic and the Broader AI Ecosystem

OpenAI, the creator of ChatGPT, and Anthropic, known for Claude, are among the first high-profile firms to submit their registrations. Both companies have publicly affirmed their commitment to compliance, noting that their existing internal safety reviews already align with many of the law's requirements.

For smaller startups, the law presents both a compliance burden and a potential competitive advantage. Firms that can demonstrate rigorous safety practices may gain trust among enterprise customers and public sector buyers, while those that lag could face fines or be barred from operating in the state.

Industry analysts predict that New York's model could inspire similar legislation in other states, potentially leading to a fragmented regulatory landscape. Companies operating nationally will likely need to adopt a unified compliance framework that satisfies the strictest jurisdiction.

What Comes Next: Enforcement Timeline and Future Steps

The DCWP has announced a phased enforcement schedule. From October 1 to December 31, 2024, the agency will focus on reviewing registrations and safety plans, issuing advisory notices rather than fines. Starting January 1, 2025, the department will begin imposing penalties for missed filings or late incident reports.

Developers can appeal any penalty within 30 days of issuance, and the law provides for a hearing before an administrative law judge. The DCWP also plans to publish an annual transparency report summarizing compliance rates, the number of critical incidents reported, and any enforcement actions taken.

Stakeholders are watching closely for guidance documents the agency intends to release in the coming weeks. These will clarify definitions such as "critical incident" and outline acceptable evidence formats for safety plans.

Practical Takeaways for Companies and Consumers

For AI firms, the immediate priority is to complete the registration portal, verify that safety documentation meets the law's criteria, and establish an internal incident-response team capable of reporting within the 72-hour window. Legal counsel familiar with state technology regulation should be consulted to avoid inadvertent violations.

Consumers in New York can expect greater visibility into how AI systems they use are tested and monitored. The law requires companies to make a public summary of their safety plan available on their websites, giving users a clearer picture of safeguards against bias, misinformation and other harms.

As the first major AI safety law to take effect in the United States, New York's approach may set a benchmark for responsible AI development, balancing innovation with public protection.