Frontier-Model Misuse, Governance, and Agent Safety Drive Scrutiny

Reports that the FTC is investigating OpenAI, Anthropic, and other major labs remain poorly documented. OpenAI reportedly withheld GPT-6.1 Astra after failures involving scope, authorization, and activity reporting while promoting always-on agents; safety-staff departures, voluntary-regulation messaging, and international AI-policy initiatives add governance pressure but no independently verified incident evidence. A final GSA LLM procurement rule adds concrete pressure through narrowed applicability, NIST-based supply-chain flow-down, commercial-term precedence, and contractor IP protections.