Attorneys general from several US states and territories have urged congressional leaders to adopt binding AI safety rules. Their joint letter is dated September 23, 2026. California and Connecticut issued their own announcements the following day.
The signatories want government-supervised safety tests and incident investigations with public findings. Investigators would need access to records. They also call for safety leaders who can make decisions independently of short-term profit goals.
The officials seek international cooperation on the development of highly capable AI while preserving states' protective powers. Rob Bonta and William Tong cite reported safety incidents and warnings from within the industry.
The letter itself creates neither a new regulator nor new legal duties. We document it as a Risk Signal. The attacks and risk forecasts it cites do not automatically become new, independently confirmed incidents.