No new weekly standard was identified. The urgent extension is runtime guidance for agent identity, permissions, tool calls, evidence, interruption and recovery.
US Government
Federal attention converged on frontier safety, protection of model intellectual property and prevention of AI-enabled cyberattacks.
State Government
California advanced specific audit, biological-risk and child-safety controls; San Francisco increased pressure around harmful AI advertising.
Law Enforcement
Clearview’s InquiryIQ was an adjacent-window follow-up, not part of the September 3–9 ranking. It highlights provenance, probable-cause and human-verification risks.
Global Regulation and Global Government
The EU incident review, California audit laws and U.S. federal proposal all point from voluntary principles toward testing, disclosure and enforceable accountability.
Control model for public-sector agents
Before action
During action
After action
Named owner; approved tools; scoped credentials; authorized data; scenario tests.
Authenticated identity; network policy; rate limits; human gates; interrupt path.
