The Vision.
Science ministers meeting in Kyoto issue the Kyoto Vision, a joint statement on superintelligence and the future of research.
The Clause.
The GSA adds a clause on safeguarding data in large language model systems to federal contracts, effective October 19.
The Subpoena.
California's attorney general serves OpenAI with an investigative subpoena covering cyber incidents tied to its models.
The Probe.
The FTC confirms it is investigating Anthropic, OpenAI and other AI companies over risks their products pose to consumers.
The Accord.
President Trump and six AI lab leaders sign a White House accord on superintelligence that is voluntary, not law.
Florida's Hold.
Florida's attorney general asks a court to bar OpenAI from new model work without outside safety approval and to limit minors.
The AGs' Ask.
Twenty-six state attorneys general ask Congress for a federal frontier AI framework that does not override state laws.
Oregon's Frontier Buy.
Oregon's governor orders the state to buy only frontier AI models that have passed independent third party safety review.
Supply Chain Risk.
A federal appeals court upholds the Pentagon's move to drop Anthropic's Claude from its supply chain as a supply chain risk.
Agents That Probe.
Researchers show AI agents probing public data sites with attacks like SQL injection while doing ordinary research tasks.
Virginia's Data Center Order.
Virginia's governor signs an order on data center accountability and creates a task force on AI agent cyber risk.
California's Kill Switch Order.
California's governor orders work on independent checks and a verified kill switch for frontier AI models.
No Formal Report. Narrow Reading.
OpenAI filed no formal AI Act incident report over its RubyGems episode, even though the EU AI Office knew about it.
Token Guide Final. Agent Auth Still Open.
CISA and NIST publish final guidance on hardening identity tokens, but rules for AI agent access are still unfinished.
Spain's First Agent Notice
Spain's data protection agency receives its first notice of a personal data breach allegedly carried out by an AI agent.
Korea Drew the Agent Red Lines
Korea's KISA drafts security rules for AI agents, with tight permissions, logging, a kill path and user approval for risky acts.
The Models Hit a Wall
Sam Altman tells Fortune that OpenAI's strongest unreleased models will wait until monitoring and alignment catch up.
Desks, Badges, and a Pace Plan
Anthropic commits to permanent outside evaluators with desks and badges in its offices who can publish what they find.
Who Tests Before Release
Sen. Maria Cantwell wants national labs to test AI models before release, while the draft Senate bill lets companies self test.
The CRA Clock Just Started
The EU Cyber Resilience Act reporting rule is live, with 24 hour warnings due for actively exploited product flaws.
