The abrupt U.S. shutdown of Anthropic’s most advanced AI systems has given way to a cautious reopening, as Washington tests how far it can restrict powerful models without crippling the very cyber defenses it says it wants to protect.

Over June 12–26, the Commerce Department’s export-control order forced Anthropic to disable both its cybersecurity model Mythos 5 and its more general-purpose sibling Fable 5, after officials warned that safeguards could be bypassed.¹ The move stunned the industry, with one outlet noting that the government’s “abrupt removal of the AI models” created uncertainty among allies whose cyber capabilities were suddenly “stunted by the loss.”²

By June 26, the Trump administration began to soften its stance. Commerce Secretary Howard Lutnick wrote that he had “determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,” clearing Anthropic to restore Mythos 5 for more than 100 specified U.S. companies and agencies, including foreign-national staff and Anthropic’s own non‑U.S. employees.³ Tech outlets framed the shift as the government “releas[ing] Anthropic Mythos to be used by more than 100 US companies, agencies.”⁴

Anthropic publicly confirmed that Mythos 5, “our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure,” adding that it was working “to expand access to Mythos 5 and make Fable 5 available for general use again.”¹ One report summed it up: “Anthropic’s Mythos 5 is back” for “a small group of cyber defenders and infrastructure providers.”⁵ Another called it a “limited carveout from US restrictions.”¹

Yet Fable 5 remains offline. Officials have kept all export controls in place for organizations not named in the government’s annex, and explicitly “did not address the release of Fable 5 in this directive.”³ Axios reported that “powerful Anthropic model, Fable 5, [is] on track to return soon,” with insiders expecting the limits “could be lifted as soon as this coming week,” pending Pentagon and NSA approval.⁶

Developers and rival labs see a broader, unsettled regime emerging. The Financial Times observed that while the administration’s partial rollback “allows some access to Anthropic’s Mythos,” it leaves “unease over Washington’s ad hoc regulatory approach.”⁷ A parallel case saw OpenAI’s GPT‑5.6 similarly constrained to a “small group of trusted partners,” prompting OpenAI to warn that such a process “should not become the long-term default” because it “keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”¹

Despite the chill, some in the ecosystem are already adapting as they wait for Fable 5’s return. One AI leader amplified a post noting that, “While we eagerly await Fable 5's return, our agentic WebGPU kernel optimization framework kept running,” with another model “pick[ing] up where Fable left off” in the browser.clem 🤗 (ClementDelangue)⁸

For now, Mythos 5’s partial comeback signals a tentative compromise: the U.S. keeps a tight grip on frontier AI models while conceding that, in cybersecurity at least, shutting them down entirely may create new risks of its own.