Reauthorized
The return of Fable 5 ends an eighteen-day export ban, while California didn’t wait for Washington’s approval to sign its own deal with Anthropic.
Fable 5 launched June 9, drew an export control ban three days later, and returned to global availability yesterday after Anthropic persuaded the Commerce Department that the jailbreak behind the ban worked on every other frontier model too. The same week, California signed the largest state-level AI deployment in the country with the same company that the Pentagon still formally considers a supply-chain risk. The federal executive order meant to establish a standing review process for frontier AI models reaches its first set of deadlines today, but the evaluation framework it envisions remains a month from completion.
False alarm
On June 12, the Commerce Department ordered Anthropic to block all foreign nationals from accessing Fable 5 and its less restricted sibling Mythos 5, citing national security concerns after Amazon researchers reported a jailbreak technique that prompted the model to identify software vulnerabilities. The directive extended to Anthropic’s own non-citizen employees, and because the company had no mechanism to verify user nationality in real time, it disabled both models globally.
Anthropic spent the following two weeks demonstrating that the same exploit produced comparable results on Claude Opus 4.8, OpenAI’s GPT-5.5, and Moonshot’s Kimi K2.7, framing the jailbreak as a general property of frontier models rather than a capability unique to Fable 5. Co-founder Tom Brown replaced CEO Dario Amodei as lead negotiator with the White House, a shift that multiple reports credited with moving the talks from political confrontation to technical review. Amodei had drawn administration hostility both for his outspoken safety positions and for his support of Kamala Harris in 2024. Commerce Secretary Howard Lutnick withdrew the export controls on June 30, addressing his letter to Brown directly, and Fable 5 returned to global availability yesterday, eighteen days after it went dark.
California goes all-in
One day before the export controls came down, Governor Gavin Newsom announced a partnership making Claude the first AI productivity tool available to every California state agency and local government, at a 50% discount with free workforce training from Anthropic engineers. The deal formalized a patchwork of existing deployments: the DMV already used Claude for customer service, the Department of Healthcare Services applied it to Medicaid caseworker workflows, CalOES and the state technology department used Claude Security and Claude Code for cybersecurity scanning and patching, and a deliberative democracy platform called Engaged California ran on Claude infrastructure.
California’s CIO and technology director Chris Given told Politico that the Pentagon’s supply-chain risk designation “just didn’t come up” during contract negotiations. A federal judge had already blocked that designation, ruling that it served to punish Anthropic for refusing the Pentagon’s contract terms rather than to protect national security. The state built its own procurement framework through a March executive order requiring AI vendors to demonstrate responsible practices on bias prevention and civil rights, creating a parallel governance track that did not depend on federal timelines or approvals.
Flying blind
President Trump signed an executive order on June 2 titled “Promoting Advanced Artificial Intelligence Innovation and Security,” setting 30-day deadlines for operational measures including a CISA binding directive on federal cyber defense and an interagency AI vulnerability clearinghouse. Those deadlines arrive today with no public announcements of completion. The larger deliverable, a classified benchmarking process and voluntary pre-release evaluation framework for covered frontier models, carries a 60-day deadline that falls on August 1.