Federal and state AI rules diverge
Governments filed several concrete AI policy actions yesterday. The Trump administration is reportedly drafting an executive order that would require formal vetting of advanced models before release, a departure from prior deregulation efforts. The move follows reports of security issues in models such as Anthropic's Claude Mythos.
USDA audit flags missed deadlines
The Department of Agriculture's Office of Inspector General issued a report documenting that the agency failed to meet Office of Management and Budget deadlines for AI governance requirements. The audit recommended tighter risk reviews and compliance procedures for the department's existing AI systems. This is one of the first public federal audits to measure actual adherence to earlier governance mandates.
Colorado narrows its AI law
Colorado Governor Jared Polis signed Senate Bill 26-189, repealing the state's earlier broad risk-based AI statute and replacing it with a narrower transparency regime. The new law, effective January 1, 2027, focuses on disclosure obligations for automated decision systems and carves out exceptions for HIPAA-covered entities. Several states continue to adjust their approaches after initial comprehensive frameworks proved difficult to implement.
California restricts AI therapy claims
The California Senate passed Senate Bill 903, which bars AI systems from providing or advertising mental health therapy. The bill requires human professionals to maintain responsibility for treatment and prohibits non-human entities from making direct therapeutic recommendations. It addresses documented cases of AI chatbots offering clinical advice without access to patient context or accountability structures.
EU extends high-risk deadlines
European negotiators reached a provisional agreement on amendments to the EU AI Act. The package postpones compliance dates for high-risk systems by roughly a year and adds new prohibitions on AI systems that generate non-consensual intimate imagery or child sexual abuse material, effective December 2, 2026. The changes also launch a public consultation on high-risk classification guidelines.
Illinois advances safety and whistleblower bill
Illinois Senator Mary Edly-Allen advanced Senate Bill 315, which would require large AI developers to assess catastrophic risks, report incidents, and conduct third-party audits. The measure includes strengthened whistleblower protections for employees who disclose safety concerns. Similar proposals have appeared in other states but remain pending.
The record will show that regulators are still choosing between tightening controls on frontier systems and scaling back earlier broad mandates.
