The Trump administration lifted its 18-day export-license requirement on Anthropic’s Claude Mythos 5 and Fable 5 on the evening of June 30, and Anthropic began restoring access July 1. The reversal landed exactly one day after Gov. Gavin Newsom announced that California would become the first state to deploy Claude across its agencies at a 50% discount.

Commerce imposed the license requirement on June 12, citing the Mythos-class models’ capacity to identify and exploit software vulnerabilities. Commerce Secretary Howard Lutnick said Anthropic had agreed to proactively detect security risks, coordinate on release protocols, and inform the government of malicious activity. Anthropic says a new safeguard blocks the jailbreak officials were worried about 99% of the time; Commerce’s Center for AI Standards and Innovation tested the guardrails.

The choreography rhymes with earlier maneuvers. Last week, Lutnick had already cleared Mythos 5 for release to a select group of White House–approved customers, the same channel OpenAI’s GPT-5.6 rolled out through. The August deadline in Trump’s June executive order to publish standardized frontier-model benchmarks is now the governing document these one-off approvals are backfilling.

Sacramento isn’t waiting. Newsom’s deal, negotiated through the California Department of Technology and its SITeS portal, extends Claude, Claude Security, and Claude Code to CalOES, the DMV, DHCS, and local governments. Asked about Anthropic’s Pentagon “supply-chain risk” designation earlier this year, California CIO Chris Given told POLITICO it “just didn’t come up” in negotiations.

That sentence is the story. A frontier lab spent 18 days on a federal cybersecurity leash while the country’s largest state was quietly writing it into permanent procurement. The federal-state rift on AI governance now runs through the same vendor’s contract portal.

Sources

Sources