Fable 5 and Mythos 5 Are Now Offline Globally. Anthropic Says the Government's Jailbreak Evidence Is Verbal and Narrow.

Since Anthropic confirmed the models were offline earlier this week, the situation has clarified in one important way: as of June 13, users worldwide are still unable to access Fable 5 or Mythos 5, according to TIME. This is not a targeted foreign-access block. It is a full global shutdown.

The reason is practical. The Trump administration's export-control directive ordered Anthropic to suspend access for all foreign nationals, including Anthropic's own foreign-national employees, according to TIME's reporting. Anthropic said the only way to comply was to "abruptly disable" the models for every customer, everywhere.

What the government actually told Anthropic

The directive came with almost no documentation. Anthropic said it received only "verbal evidence of a potential narrow, non-universal jailbreak", according to The Guardian. No written technical finding. No specific details of the national security concern, the company said.

The alleged vulnerability: a method of bypassing safeguards that would allow Fable 5 to be used to identify software vulnerabilities, according to The Guardian. That is a serious concern if real. Offensive use of frontier AI to hunt for exploitable code flaws could give a foreign adversary a meaningful edge against U.S. infrastructure.

But Anthropic's public statement draws a sharp line: "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people."

This is a first in U.S. export-control history

For years, U.S. export controls targeted the hardware layer — chips, fabrication equipment, design software. Restricting access to the AI model itself is a different category entirely, and TIME notes this order marks the first time the government has used export controls this way.

If the government can force an offline shutdown of a commercial AI model on the basis of verbal, unwritten evidence of a potential vulnerability, every major AI lab now has a new kind of regulatory exposure that has no established legal framework around it.

The case for the government's position

If Fable 5 can be reliably jailbroken to function as a vulnerability-discovery engine, the risk is not theoretical. Foreign intelligence services — China's in particular — have documented histories of using commercial Western technology for offensive cyber operations. A verbal briefing may reflect the sensitivity of the underlying intelligence. Requiring the government to produce written technical evidence before acting could itself be a security risk if the evidence comes from classified sources or ongoing operations. Critics of Anthropic's pushback would argue the company has a massive financial interest in keeping its flagship models online ahead of its IPO and should not be the final arbiter of what constitutes an acceptable national security risk.

But "the government said so verbally" is not a standard that would survive scrutiny in any other regulated industry. The FDA does not pull a drug on verbal evidence. The FAA does not ground a fleet on a verbal report. The absence of a written finding makes independent verification impossible, and it makes Anthropic's own engineers unable to confirm whether the jailbreak is real, narrow, or already patched.

The political backdrop

This shutdown does not happen in a vacuum. As The Guardian reported, Anthropic's relationship with the Trump administration had already fractured earlier this year after the company refused to allow the U.S. military to use its models for domestic surveillance and fully autonomous weapons systems. The government responded by placing Anthropic on a supply chain blacklist, set to take effect later in 2026.

That context does not prove the jailbreak finding is fabricated or pretextual. But it does mean the government and Anthropic are not neutral parties working through a routine security review. They are adversaries in an ongoing dispute, and both sides have stated interests that readers should factor in.

The Guardian also noted that as recently as Wednesday, June 11, Anthropic had publicly called for greater U.S. government oversight of AI, including the power to block models with unacceptable risks. The company said the Friday directive "did not follow principles of fair and fact-based regulation." That is a precise objection: not that oversight is wrong, but that this specific action skipped the evidentiary steps that make oversight legitimate.

Pentagon Chief Information Officer Kirsten Davies posted publicly in connection with the order, according to The Guardian, though the source text was truncated before her full statement was captured. Her exact position on the record remains unclear from available sourcing.

What comes next

Anthropic said it is "working to restore access as soon as possible" and believes the situation is "a misunderstanding," according to TIME. That framing suggests the company expects a negotiated resolution rather than a legal fight — at least for now.

The unresolved question is whether the government will produce written, verifiable evidence of the jailbreak. If it does, Anthropic's public resistance looks reckless. If it does not, the shutdown stands as a precedent: a government can take a commercial AI product used by hundreds of millions of people offline, globally, with no documented technical finding, and the burden falls on the company to prove the concern is wrong rather than on the government to prove it is right.