Artificial intelligence models: large developers.

Senator Wiener's comprehensive artificial intelligence legislation establishes new oversight requirements for large AI developers in California while creating a public computing infrastructure to advance ethical AI development. The Transparency in Frontier Artificial Intelligence Act requires developers of foundation models to publish detailed safety protocols, undergo annual third-party audits, and report critical safety incidents to the Attorney General.

The legislation defines foundation models as AI systems trained on broad datasets that can adapt to various tasks. Large developers - those using substantial computing power and earning over $100 million annually - must document their testing procedures for assessing catastrophic risks, including potential malfunctions or loss of control that could cause significant harm. Starting in 2030, these developers must retain independent auditors to evaluate their compliance with safety protocols and submit summary reports to state authorities.

A new consortium within the Government Operations Agency will develop a framework for CalCompute, a public cloud computing cluster aimed at expanding access to AI research capabilities. The 14-member group, including representatives from academia, labor organizations, and technical experts, must submit implementation recommendations by 2027 addressing governance, costs, and workforce development.

The bill also strengthens protections for employees who report potential catastrophic risks or violations by large AI developers. Companies must establish anonymous reporting channels and cannot retaliate against workers who disclose safety concerns to management or government authorities. The Attorney General can impose graduated civil penalties ranging from $10,000 to $10 million for violations, with enforcement provisions allowing successful plaintiffs to recover attorney fees.