State AI Safety Laws: California and New York
The landscape of artificial intelligence regulation is evolving rapidly, particularly with the recent enactment of AI safety laws in California and New York. These states are taking significant steps to address the potential risks associated with large AI models, aiming to create a safer environment for deployment.
Landmark Convergence
California’s Transparency in Frontier Artificial Intelligence Act (TFAIA) was the first state law to address potentially catastrophic risks from AI. Notably, the New York governor adopted much of California’s provisions to establish a “unified benchmark” for AI regulation across states. This convergence highlights the need for a cohesive approach to AI safety.
Large Developers and Models
Currently, only a limited number of companies and models meet the revenue and compute thresholds established by these laws. These thresholds will undergo annual evaluations and may be revised based on technological advancements and evolving standards.
Unclear Future
Given the recent Executive Order 14365 issued by the White House, the future of state AI laws remains uncertain. The order aims to formulate a national policy framework for AI laws while minimizing regulatory fragmentation. It is unclear whether California and New York’s AI safety laws will face challenges under this executive order.
Key Provisions of the AI Safety Laws
Both California and New York have established frameworks to impose protections for the safe deployment of large “frontier AI models.” Below are key provisions comparing the two laws:
- Core Requirement: Requires a documented technical and organizational framework for managing risks — similar requirements exist in both states.
- Models: Covers frontier models defined by specific operational thresholds, with similar definitions utilized.
- Developers: Defines developers, including large frontier developers with a revenue threshold; similar definitions apply.
- Risk Definition: Defines catastrophic risk concerning significant harm or damage; the same risk definitions apply.
- Incident Reporting: California requires reporting critical incidents within 15 days, while New York requires reporting within 72 hours.
- Public Disclosure: Developers must publish transparency reports; this requirement exists in both states.
- Penalties: Penalties can reach up to $1 million per violation in California and can be as high as $3 million for subsequent violations in New York.
- Whistleblower Protections: California protects employees responsible for risk assessment; New York does not include similar protections.
Conclusion
The enactment of the TFAIA in California and the RAISE Act in New York marks a pivotal moment in the regulation of artificial intelligence. These laws not only aim to mitigate catastrophic risks associated with frontier AI models but also set a precedent for other states to follow. As these regulations unfold, their impact on innovation and safety in AI deployment will be closely monitored.
