Commerce Secretary Howard Lutnick announced at 11:37 p.m. Eastern on June 30 that the Trump administration was lifting the export controls that had forced Anthropic to disable Claude Fable 5 and Claude Mythos 5 worldwide for 18 days. Fable 5 returns to Claude.AI and Claude Code on Wednesday, along with distribution through Amazon Web Services, Google Cloud, and Microsoft Foundry.

The reversal closes an episode that started at 5:21 p.m. Eastern on June 12, when a directive citing national security authorities landed at Anthropic with no specific rationale attached. Hundreds of millions of users lost access to the company’s frontier models. On June 26, Mythos 5, the more capable underlying model that Fable 5 is a guardrailed public version of, was quietly cleared for certain U.S. organizations through Anthropic’s Glasswing cybersecurity program. Four days later, the broader ban dissolved.

“Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the U.S. Government and strengthen America’s leadership in A.I.,” Lutnick wrote on X. In a letter to co-founder Tom Brown viewed by CNBC, he told the company that “appropriate safeguards” were in place to permit certain “trusted partners” to access the model.

The framing is worth reading closely. What Lutnick describes as collaborative analysis is, functionally, a bespoke federal vetting process invented on the fly for a single company’s model release. There’s no statute defining it, no published criteria, no timeline. There’s a directive, a silence, and eventually a letter.

Anthropic has been sideways with the administration for months. In March the company sued the Defense Department, which had labeled it a “supply chain risk.” Dario Amodei, the chief executive, publicly backed Kamala Harris in the 2024 election, and has since been a recurring target of White House criticism. That the ban arrived without a stated reason, and that its lifting arrived with a warm X post, suggests the operative variable was never a technical safety finding.

Competitors noticed. OpenAI, whose upcoming GPT-5.6 the White House has reportedly asked to restrict to government-approved partners, said in a statement that “We don’t believe this kind of government access process should become the long-term default.” It rarely does, until it does. The 1996 Telecommunications Act was also drafted as a one-time settlement of a specific fight, and structured American communications policy for a generation. What began as an 18-day standoff over one company’s model has quietly established that frontier releases in the United States now clear at the pleasure of the Commerce Secretary.

Sources