Claude Fable 5 and Mythos 5 Are Back Online: Inside the 18-Day Government Restriction That Reshaped How America Regulates Frontier AI

Eighteen days after the Trump administration abruptly forced Anthropic to cut off global access to its two most powerful AI models, Commerce Secretary Howard Lutnick signed a letter on Tuesday night lifting all export controls — and Anthropic began restoring access on Wednesday morning. The episode, short in duration but consequential in precedent, has left the AI industry navigating a newly visible and still largely undefined regulatory relationship between Washington and the companies building the world's most capable artificial intelligence systems.

How the Restriction Began: A Jailbreak, a Shutdown, and No Clear Explanation

On June 12, 2026, the Trump administration ordered Anthropic to immediately cut off access to Claude Fable 5 and Claude Mythos 5 — its most advanced publicly available and enterprise models — for all foreign nationals, including Anthropic's own non-US-citizen employees. Anthropic complied immediately, what it described in a public statement as an "abrupt" disabling of both models for all customers.

At the time, Anthropic said the government had not provided a specific reason for the order beyond unspecified national security concerns, but that it believed officials were responding to a reported "jailbreak" of Fable 5 — a method researchers had reportedly demonstrated for circumventing the model's built-in safety guardrails. Anthropic immediately pushed back on the framing, saying the technique had been used only to identify "a small number of previously known, minor vulnerabilities" rather than to enable any sophisticated cyberattack, and characterizing the shutdown as disproportionate to the actual risk demonstrated.

The restrictions came against a particularly fraught backdrop. Just three months earlier, in March 2026, the Pentagon had labeled Anthropic an "unacceptable supply chain risk" following tense negotiations over the company's refusal to provide assurances its models would not be used for mass surveillance or autonomous weapons. Anthropic sued the Department of Defense over that designation — making June's abrupt model shutdown the second major confrontation between the company and the federal government within a single quarter.

The 18-Day Process: Partial Restoration, Then Full Clearance

The resolution came in two stages. On June 26 — 14 days after the initial shutdown — the administration granted Anthropic permission to restore Mythos 5 access specifically for American organizations working in critical infrastructure cybersecurity, a narrow and explicitly security-focused restoration that sources told Axios was also a signal that a broader lifting was coming soon. Four days later, on June 30, Lutnick's letter formally reversed the June 12 order entirely.

"Anthropic has taken steps in close coordination with the U.S. government to address the risks associated with Claude Mythos 5 and Claude Fable 5," Lutnick wrote in the letter, which was widely circulated online. The Commerce Department's Center for AI Standards and Innovation independently tested the model's previous and updated safeguards. The key technical outcome: Anthropic said it had implemented a new safeguard that blocks the specific jailbreak that had triggered the original government concern 99% of the time, and that in the remaining 1% of cases, the model's output included only previously discovered or already-patched vulnerabilities — not new, exploitable security flaws.

Lutnick posted on X that his office had "worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI." White House Chief of Staff Susie Wiles described the resolution as evidence of government and private sector working together "in a way we have never seen before."

The Conditions Attached: Ongoing Obligations, Reservation of Rights

The restoration is not unconditional. In his letter to Anthropic, Lutnick set out three ongoing commitments the company must maintain to keep its models accessible: proactively detecting and addressing security risks associated with the models, working with the US government on protocols and standards for current and future models, and informing the government of any malicious activity identified in connection with the systems.

Lutnick also included an explicit reservation of rights, warning that the Trump administration reserves the right to reimpose restrictions in the future if "circumstances change or should Anthropic fail to adhere to its commitments." That clause — a hanging condition over Anthropic's ongoing ability to serve its global customers — captures the structural novelty of what has now been established: a relationship in which a private AI company's most advanced models can be taken off the market by executive action and returned by executive approval, with the company's continued access dependent on an ongoing compliance relationship with the Commerce Department.

A Broader Pattern: The Government's New Role in Frontier AI

The Anthropic episode does not stand alone. It is the most dramatic single instance of a pattern that is now clearly visible across the frontier AI industry. When OpenAI unveiled GPT-5.6 Sol in June, the administration requested that the company limit its rollout to a small group of government-approved partners before broader public release — a request OpenAI complied with, rolling out the model to what it called "a small group of trusted partners" rather than making it immediately universally available.

The context is a June 2026 executive order signed by President Trump asking technology companies to voluntarily give the government oversight of new AI models before releasing them to the public — the administration's most significant step yet toward a formal regulatory posture on AI, following its earlier laissez-faire approach at the start of the second term. The Commerce Department is operating under an August deadline established by that executive order to create standardized benchmarks for evaluating the security risks of new AI models — a deadline that gives the current ad hoc, case-by-case approach a defined sunset and the promise of something more systematic in its place.

Francesco Bailo, deputy director of the AI, Trust and Governance Centre at the University of Sydney, told Al Jazeera the lifting of restrictions had been expected in technical circles once the scale of the jailbreak risk was properly assessed. "Reports that researchers were able to 'jailbreak' Fable 5 had been widely inflated beyond their actual significance," Bailo said — a characterization that aligns with Anthropic's own public assessment of the vulnerability throughout the episode.

The Question the Industry Is Now Asking

Tanishq Abraham, former research director at Stability AI and now CEO of medical AI company Sophont, framed the precedent question most directly. "The biggest question now is: What precedent does this set for the industry? Does the US government need to approve every frontier model release?" Abraham told Al Jazeera. He attributed the improving relationship between Anthropic and the government in part to the efforts of Tom Brown, an Anthropic co-founder who has been engaged in navigating the company's relationship with Washington.

That question — whether executive-branch review of frontier model releases is becoming a permanent feature of the American AI regulatory landscape, rather than a temporary emergency measure — is the one that will define the months between now and the administration's August benchmarking deadline, and beyond. The Anthropic episode has made visible something that was previously implicit: that the US government now considers itself a stakeholder in the deployment decisions of the country's most advanced AI systems, with the authority to restrict access on national security grounds and the expectation of ongoing consultation as new, more capable models are developed.

Anthropic said it was grateful for its users' patience and thanked "everyone who worked with us on redeploying the models." Claude Fable 5 and Mythos 5 are accessible again as of Wednesday, July 1. The next frontier model release — by Anthropic, OpenAI, Google, or any of the other companies building at the edge of AI capability — will arrive in a regulatory environment that looks meaningfully different from the one that existed three weeks ago.

Sources

1. Al Jazeera — US lifts restrictions on Anthropic's powerful AI models Fable and Mythos
2. Axios — Trump administration lifts restrictions on Anthropic's Fable 5
3. Forbes — White House Lifts Restrictions On Anthropic's Mythos And Fable AI Models
4. The Washington Post — White House drops restrictions on Anthropic AI models after two-week ban
5. GV Wire / The New York Times — US Lifts Restrictions on Anthropic's Most Powerful AI Models
6. Euronews — US lifts export controls on powerful AI models, Anthropic says