Scared of artificial intelligence? New law forces makers to disclose disaster plans

Scared of artificial intelligence? New law forces makers to disclose disaster plans

https://laist.com/news/new-law-forces-ai-makers-to-disclose-disaster-plans

Publish Date: 2025-12-31 15:00:00

Source Domain: laist.com

• Whistleblower Protections: Governor Gavin Newsom signed a law beginning January 1, 2024, that provides whistleblower protections to employees in large tech companies working on AI models, allowing them to report safety incidents if concerns arise that things may go wrong.

• Framework Requirements: The law mandates that companies developing large AI models must publish frameworks detailing how they respond to critical safety incidents, assess, and manage catastrophic risks. Companies failing to comply may face fines up to $1 million per violation.

• Catastrophic Risk Definition: The law defines a catastrophic risk as an AI incident that kills more than 50 people through cyber attacks or hazardous weapons, or results in over $1 billion in theft or damage, focusing on instances where control of the AI system is lost.

• Public Transparency Measures: Large AI model developers are required to share more information on their operations with the public. This includes reporting critical safety incidents to the state within certain timeframes and publishing transparency reports detailing intended model uses, risk assessments, and third-party reviews.

• Criticisms on Limitation: Critics argue the law is incomplete. It does not consider environmental impacts, disinformation issues, or perpetuation of oppression by AI, nor does it apply to government-developed AI profiling systems. Additionally, transparency measures fall short as incident reports to the Office of Emergency Services are not fully accessible to the public.

• Assembly Bill 2013: Effective January 1, 2024, Assembly Bill 2013 requires companies to disclose more details on the data used to train AI models, potentially adding more transparency to the AI training processes.

• Future Reporting: Starting in 2027, the Office of Emergency Services will produce an anonymized report on critical safety incidents from the public and large frontier model creators, though specifics on which AI models are involved will be withheld from the public.