Anthropic got a piece of its frontier AI back on Friday, and not much more. The US government told the company that Mythos 5, its most powerful cybersecurity model, can be switched on again. But the win is thinner than it sounds. Mythos 5 is only being made available to a vetted set of roughly 100 organizations that defend critical infrastructure. Fable 5, the public-facing model that millions briefly got to use before it vanished, is still dark with no timeline for return.The export control directive that knocked the models offline two weeks ago, the one that triggered a global outcry over a government deciding who gets to use frontier AI, has not been lifted. The foreign-national ban that locked out even Anthropic’s own employees still stands for anyone off the approved list. And the deeper grievance—that Washington now decides, company by company, who gets the best AI tools—hasn’t gone away. What changed is that the Commerce Department carved out an exception, and called it progress.
Why Anthropic’s “restored” doesn’t mean what it sounds like
Commerce Secretary Howard Lutnick’s letter, dated June 26 and addressed to Anthropic co-founder Tom Brown, framed it as a “revision to the license requirements.” Anthropic, he wrote, had “worked with the U.S. government to address risks associated with the covered models,” and those efforts “yielded significant progress.” On that basis, Lutnick determined “appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.“Read closely, and the limits are everywhere. Lutnick added that “all other requirements of the June 12 letter remain in effect until further notice” and reserved the right to adjust the rules again. The foreign-national ban—which barred even Anthropic’s own non-US employees from touching the models—stays, except for approved companies and Anthropic staff working on the cleared list. This isn’t a door reopening. It’s a window cracked for a hand-picked few.
Fable 5 stays dark while Mythos comes back
Here’s the part that matters to ordinary users: Fable 5 is still dark. It’s the same underlying model as Mythos, but tuned with safeguards for general release, and it was the first time Anthropic put something this capable in front of the public—for about three days, before it vanished on June 12. The Friday letter says nothing about bringing it back. Negotiations were expected to run into the weekend with an eye toward Fable, a source told CNN, but there’s no timeline and no agreement. So the model the world actually got to use remains the one the world can’t have.The whole mess traces back to a jailbreak. Anthropic says the government acted on research—reportedly from Amazon engineers—showing Fable’s guardrails could be coaxed into flagging software vulnerabilities. Anthropic pushed back hard, arguing the demonstrated capability was narrow, involved already-known minor flaws, and was “widely available from other models” including OpenAI‘s GPT-5.5. Pulling a model used by hundreds of millions over that, the company said, would “essentially halt all new model deployments” if applied industry-wide.
Altman, Anthropic and the customer-picking problem
The deal de-escalates a genuinely strange standoff, but it hands critics a fresh complaint: now the government is picking customers. “No one knows how these companies are picked and why everyone else is excluded,” said John Coleman of the Foundation for Individual Rights and Expression, warning it “raises questions about the rule of law.” Even Sam Altman, whose OpenAI just agreed to its own staggered, government-approved rollout of GPT-5.6, said the safety testing is fine but “I just don’t like the idea of the government picking the customers.“That parallel is the wider story. On the same Friday, OpenAI limited its new GPT-5.6 lineup to a small group of vetted partners at Washington’s request, calling it a “short-term step” it doesn’t want to become “the long-term default.” Two of the biggest US labs, both now releasing frontier models on the government’s terms, both grumbling publicly while complying.Anthropic’s road there was rougher than most. The administration has called Amodei an “ideological lunatic,” labeled the company a “supply chain risk” after it refused to let the Pentagon use its models for domestic surveillance and autonomous weapons, and Anthropic has sued over that blacklisting. Friday’s letter thaws one corner of a relationship that’s been frozen for months.For now, a hundred-odd organizations get Mythos 5 back. Everyone else—including the public Fable was built for—waits on US governemnt’s discretion. The ban the world protested is still on the books. A small group just got an exception.
