Anthropic has won back broad access to Claude Fable 5 after an 18-day fight with Washington, but Mythos 5 is still moving through a narrower gate.
The Department of Commerce has lifted the emergency controls that knocked Anthropic's top models out of normal service, and that matters because the fight went straight at the thing investors and customers need most from an AI company: confidence that its best product can actually stay online.
The timeline was brutal. Claude Fable 5 launched publicly on June 9. By June 12, Commerce Secretary Howard Lutnick had ordered Anthropic to suspend access for foreign nationals, including foreign employees working inside the United States, according to Business Insider and the New York Post. Anthropic shut down Fable 5 and Mythos 5 worldwide rather than try to run a compliance maze around nationality, geography and model access in real time.
The trigger was a jailbreak concern, not a normal product bug. The Verge reported earlier in June that Amazon researchers had flagged red-team findings around Fable 5, after prompts pushed the model toward identifying software vulnerabilities and, in at least one case, code showing how one could be exploited. If you're running Anthropic, that is not just a safety issue. It's a sales issue, a legal issue and a Washington issue all at once.
Anthropic pushed back on the government's framing. The company argued that the cited technique was narrow and did not amount to a broad collapse of Fable 5's safeguards. It also said similar behavior could be drawn from rival models that were not hit with the same restriction. That is the part Washington still has not explained cleanly. You can regulate powerful models because they create cyber risk, or you can let the market run and accept the risk. Picking one company for the hardest treatment while competitors keep selling comparable tools is not a policy. It's pressure.
Fable 5 was too important to sit on ice. Anthropic priced the model at $10 per million input tokens and $50 per million output tokens at launch, less than half the cost of Mythos Preview, and pitched it as one of its strongest systems for software engineering, scientific research and agentic work. Axios reported on June 30 that Anthropic was already positioning the cheaper Claude Sonnet 5 as the everyday workhorse while Fable and Mythos sat under government scrutiny. That is a useful fallback. It is not a replacement for the model enterprise customers had just started building around.
Mythos 5 is a different case. It was built for cybersecurity work and released only to a smaller set of trusted users, including cyberdefenders and infrastructure operators. Business Insider reported that some U.S. organizations got a limited carveout for Mythos 5 before the broader Fable decision landed. The Guardian reported on July 1 that Mythos access is being restored more carefully to selected U.S. cybersecurity organizations, while Fable access is being restored more broadly.
That distinction matters. Anthropic did not simply get everything back as if the last 18 days were a clerical error. Fable 5 is returning to global access starting Wednesday, July 1, while Mythos 5 remains tied to approvals, monitoring and closer cooperation with the U.S. government. Lutnick said the department had worked with Anthropic on model security and American AI leadership. Strip out the ceremony and the bargain is clear enough: Anthropic gets its product back, and Washington gets a firmer hand on the most sensitive versions of it.
The timing could hardly be more awkward. The New York Post reported that Anthropic confidentially filed for an IPO on June 1 after a funding round that valued the company near $965 billion. Business Insider also reported the S-1 filing, while the company was still dealing with its wider fights with the Trump administration over military and national security use. An unresolved export-control order on a flagship model is exactly the kind of overhang that makes public-market investors ask whether the revenue line is really under management's control.
Frankly, the biggest story here is not that Anthropic got a reprieve. It is that the government has shown it is willing to reach directly into model access, not just chips, data centers or cloud contracts. That is a much bigger warning for OpenAI, Google, Meta and any startup trying to sell frontier capability into global markets. If a model can be switched off for foreign users after one reported jailbreak finding, your product roadmap is no longer just an engineering document.
Anthropic is back in motion, but the policy logic is still unfinished. If Fable 5 was dangerous enough to justify a worldwide shutdown, then similar models deserve the same scrutiny. If it was not, the last 18 days were a costly lesson in arbitrary power. Either way, the next frontier AI launch will arrive with one more audience in the room: Commerce officials who now know they can pull the brake.
Also read: Tech and Finance Are Cutting Jobs by the Thousands as AI Spending Climbs • Goldman Sachs says Europe's grid and factory suppliers are cashing in on AI • Abu Dhabi's MGX Closes Nearly $50 Billion to Bankroll the AI Buildout
