Trump Administration Pulled Anthropic's Two Most Advanced AI Models Offline Worldwide, Citing a Potential Jailbreak

Since the Pentagon designated Anthropic a supply chain risk in February 2026, the relationship between the Trump administration and the AI safety company has deteriorated steadily. Friday's export-control directive, reported by TIME, The Guardian, and Crypto Briefing, represents the most concrete and operationally disruptive step yet.

What the Order Actually Did

The directive, issued June 13, ordered Anthropic to suspend access to Fable 5 and Mythos 5 for any foreign national — including Anthropic's own non-U.S. employees. Because Anthropic had no practical way to verify nationality at the individual user level at that moment, the company shut down both models entirely for all users worldwide.

Users reported losing access Saturday. Anthropic confirmed the outage in a public statement: "We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible."

The scale matters. Fable 5 and Mythos 5 are Anthropic's most capable models, deployed to what the company described as hundreds of millions of users.

The Jailbreak Claim

The Guardian reported the specific technical concern underlying the order: the U.S. government believes there is a method to bypass, or "jailbreak," a safeguard that would otherwise prevent Fable 5 from being used to identify software vulnerabilities — the kind of capability that could accelerate cyberattacks on U.S. infrastructure.

Anthropic pushed back directly. The company said the government provided only "verbal evidence of a potential narrow, non-universal jailbreak" and that it "disagrees that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people."

Jailbreaks exist on a spectrum. A narrow jailbreak that works under specific conditions is a real risk. Whether that risk justifies a global commercial shutdown is a policy judgment, not a settled engineering fact.

The Broader Context

This did not come out of nowhere. Crypto Briefing laid out the timeline clearly: in February 2026, President Trump ordered federal agencies to stop using Anthropic's technology. The Pentagon then designated Anthropic a supply chain risk, effectively warning government contractors that doing business with Anthropic could jeopardize their own government relationships. Anthropic filed suit against the Defense Department in March 2026, arguing the blacklisting was legally questionable and counterproductive for U.S. AI competitiveness.

The root dispute: Anthropic refused to allow the U.S. military to use its models for domestic surveillance and fully autonomous weapons systems. CEO Dario Amodei cited the company's AI safety policies. Defense Secretary Pete Hegseth and President Trump have both publicly criticized Anthropic, claiming its stance jeopardizes national security.

As Crypto Briefing noted, rivals like OpenAI moved quickly to fill the vacuum, securing or expanding Pentagon contracts after Anthropic's government relationships collapsed.

The Strongest Case for the Government's Position

The administration's concern merits serious examination. If a foreign adversary — China, in particular — can access a top-tier U.S.-built AI model and use it to find exploitable vulnerabilities in American software systems, the damage could be significant and largely invisible until it's too late. The U.S. has used export controls on chips and hardware for years on exactly this logic. Extending that logic to the models themselves is a meaningful escalation, but it is not an incoherent one. Pentagon Chief Information Officer Kirsten Davies publicly defended the directive, according to The Guardian. Governments generally don't telegraph the specific intelligence that drives national security decisions, which is why Anthropic received verbal rather than detailed written evidence.

Still, "we can't tell you exactly why" is a hard standard for a company to accept when the order requires shutting off service to hundreds of millions of people globally.

What's New Since Prior Coverage

Friday's action crossed a line that prior coverage of this feud had not reached: the U.S. government actually ordered a commercial AI model offline, not just restricted in government procurement. This is the first export-control directive applied directly to an AI model rather than to the chips or tools used to build one, according to TIME.

The Guardian noted a particularly sharp irony in timing: as recently as the Wednesday before the order dropped, Anthropic had publicly called for greater U.S. oversight of AI, including government authority to block models with unacceptable risks. Anthropic said the Friday directive, however, "did not follow principles of fair and fact-based regulation."

One concrete open question as of June 14: Anthropic's IPO timeline. The company was described by The Guardian as "IPO-bound" and the government's blacklist action was set to take effect "later in the year." A global model shutdown over a disputed jailbreak claim, layered on top of an active lawsuit against the DoD, creates obvious complications for any public offering. Whether the two sides reach a resolution — and how fast — will likely determine both.