Five major labs scored poorly on public incident monitoring and shutdown plans. The FTC's 13 AI cases have focused on deceptive marketing rather than autonomous-agent behaviour.
Guidelight AI Standards assessed Anthropic, Google, OpenAI, Meta and xAI on five areas using only public information: internal monitoring, behaviour logging, procedures for halting systems after a safety incident, third-party audits and containment plans. OpenAI led with 3 out of 5, Google placed in the middle, and Meta and Anthropic shared the bottom. The assessment measures the scope of published plans and does not determine whether unpublished practices exist.[1]
Guidelight chief scientist Steven Adler said he was surprised by how little the companies had published about handling serious incidents. OpenAI said the scoring did not cover all of its internal practices. Meta and Anthropic did not say whether they have unpublished plans. State disclosure duties are expanding: California's SB 53 requires publication of frameworks for critical safety incidents, while New York's RAISE Act is due to take effect in January 2027.[1]
Forkast's review of Operation AI Comply enforcement since September 2024 found that all 13 Federal Trade Commission actions targeted deceptive marketing claims. The actions recovered about 51 million dollars, but none addressed the conduct of an autonomous agent itself. A Congressional Research Service note from July 2026 found no federal guidance dealing directly with agentic AI. The AI AGENT Act, introduced as a discussion draft in June 2026, has not been enacted.[2]
