Commerce Secretary Howard Lutnick cleared Anthropic's Mythos 5 frontier model and its public counterpart Fable 5 after the company agreed to new safeguards and standing threat reporting, while reserving the right to reimpose controls.
The US Commerce Department lifted export controls on Anthropic's Mythos 5 and Fable 5 artificial intelligence models on Friday, less than three weeks after the same agency had ordered the company to disable them, in a rapid reversal that exposes how AI export controls are functioning in practice: less as a permanent blacklist than as a conditional lever the government can pull, release, and pull again.
The reversal was confirmed in a letter from Commerce Secretary Howard Lutnick to Anthropic, seen by Reuters and CNBC, and in an Anthropic post on X stating: "We've received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring access tomorrow." Fable 5 is Anthropic's widely available public model, while Mythos 5 is a variant with some safety guardrails removed, an arrangement that put it at the center of the original national security review.
The Lutnick letter, according to Reuters, ties the reversal to specific Anthropic commitments: to "proactively detect and address security risks associated with the models," to work "diligently" with the US government on protocols for Mythos, Fable, and future models, and to inform Washington of any malicious activity. The letter also reserves the right to "re-evaluate the decisions made in this letter and the necessity of reimposing a licence requirement, should circumstances change or should Anthropic fail to adhere to its commitments." That reservation, more than the ban itself, is the active control. It converts a one-time national security finding into an open-ended compliance regime the company has to keep satisfying.
The trigger for the original June 12 order, according to Wired, was a "jailbreaking" technique that could be used to bypass Fable 5's safeguards. Anthropic added a new safeguard specifically targeting that technique. An Anthropic source told reporters the underlying vulnerabilities had been known and patched earlier, framing the new control as reinforcement rather than repair; the Commerce Department evidently did not share that view. The disagreement matters because the next dispute between the company and the government will turn on whether the post-reversal commitments are being met, not on whether a fresh vulnerability can be discovered.
A short intermediate step made the leverage structure visible. Before the full reversal, Mythos 5 had been allowed only to a small group of "trusted" US organizations, a partial release that signaled the door could be opened incrementally. The Center for Strategic and International Studies, in an analysis of the original restriction, framed the case as a stress test of how Commerce can extend export-control logic to domestic deployment of frontier models, an extension that goes beyond the traditional foreign-destination scope of US export law. The Cloud Security Alliance, in a separate research note, has begun cataloguing what the new posture means for enterprise governance, particularly for organizations that had been quietly planning around the assumption that Mythos 5 would stay restricted.
The downstream concern, repeated by several of the experts cited in the wire coverage, is that Mythos 5 in the wrong hands could accelerate sophisticated cyberattacks, especially against sectors like banking that run on complex legacy technology stacks. That is the same kind of risk the original order tried to bound, and it is the risk Lutnick's letter now asks Anthropic to help monitor in cooperation with Washington. The arrangement effectively deputizes the company as a sensor for the threat its own model enables.
The Friday reversal landed on the same day that rival OpenAI announced it was delaying a full public launch of GPT-5.6 at the US government's request. The two events are separate but temporally adjacent, and together they sketch a US policy that is choosing to slow, condition, and selectively release frontier AI rather than allow an unfettered public rollout. What to watch next is whether the Lutnick framework, conditional relief in exchange for standing threat reporting and pre-agreed reactivation authority, becomes a template for OpenAI, xAI, and Google, or whether Anthropic's three-week episode remains an isolated settlement.