A New Report Argues the US Ban on Anthropic's Models Was Never Really About a Jailbreak

--- headline: "A New Report Argues the US Ban on Anthropic's Models Was Never Really About a Jailbreak" slug: "anthropic-ban-rationale-political-questions" category: "policy" story_number: 15 date: "2026-06-19" ---

A New Report Argues the US Ban on Anthropic's Models Was Never Really About a Jailbreak

When the U.S. Commerce Department sent Anthropic a letter on the afternoon of Friday, June 12, the official rationale was straightforward: national security. An obscure export control directive barred any foreign national, "inside or outside the United States," from accessing the company's two newest and most capable models, Fable 5 and Mythos 5. Anthropic complied almost immediately, pulling both offline for every customer worldwide rather than risk running afoul of the order. But a wave of reporting that followed over the weekend argues the jailbreak that ostensibly triggered the ban was, at most, a pretext, and that the real story was about politics, leverage, and a souring relationship between the AI lab and the Trump administration.

What the order did, and how fast

The mechanics of the ban were as striking as its substance. According to TechCrunch security editor Zack Whittaker, the directive forced "a tech company to pull its models offline with a swift and unilateral action that didn't appear to require court approval." The letter itself was never made public, and it offered no specifics. Anthropic said it believed the order related to a bypass of its models' safety guardrails, but acknowledged it could not be certain because the government provided so little detail.

The sweep of the order was unusually broad. Because it barred access by any non-American, it locked out not just overseas customers but Anthropic's own foreign-national employees, leaving them unable to touch the company's flagship products. A source close to Anthropic told Fox Business the company was initially asked to take Fable 5 down within 90 minutes.

The jailbreak that "should never have triggered an export control"

The technical event behind the ban appears to be a guardrail bypass discovered after Fable 5's June 9 release. The Wall Street Journal reported the researchers who documented it were at Amazon, and administration officials told Fox Business that Amazon AI experts, along with five other companies testing the release, found a workaround that opened up the full cyber capabilities Anthropic had tried to constrain.

But the cybersecurity community pushed back hard on the idea that this warranted a federal shutdown. Katie Moussouris, founder of Luta Security and a veteran of earlier fights over export law and security research, was shown a private copy of the researchers' paper by Anthropic. In a blog post she argued the bypass amounted to little more than rephrasing a request, the difference between asking a model to "review code for security issues" and asking it to "fix this code," with largely the same result either way.

> The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense.

That is Moussouris's assessment, and she was not alone. She and dozens of other researchers signed on to call the order "dangerous," warning that pulling advanced cybersecurity capabilities away from American network defenders hurts U.S. cyber defense rather than protecting it. Anthropic, for its part, said it had reviewed a demonstration of the technique surfacing "a small number of previously known, minor vulnerabilities" that other publicly available models could also find without any bypass.

The political read

If the technical case is thin, what filled the vacuum was a political explanation. Citing sources, Axios reported that "personality differences" between Anthropic and the administration, rather than a genuine product flaw, drove the directive. Justin Hendrix, editor of Tech Policy Press, framed the broader climate bluntly: it is one "of a cloud of suspicion that senior officials are picking favorites based on personal and political factors."

The backdrop supports the suspicion. Anthropic and the Trump administration have had a fractious relationship; in March, the Pentagon labeled the company a supply-chain risk. Whittaker laid out the open questions without claiming to resolve them: Did officials misread the report and overreact? Did Amazon's leadership raise concerns out of caution, or spite? Was the order a deliberate pressure tactic against a company the White House already distrusted, with officials only later grasping how far-reaching the consequences would be? What is not in dispute is the precedent. As Whittaker put it, "This time the government took issue with Anthropic; tomorrow it could be with anyone else."

The administration's case

It would be unfair to treat the national-security rationale as obviously empty, and the administration's account deserves a fair hearing. Senior officials told Fox Business that the ban stemmed not from the bug itself but from Anthropic's "recklessness" in responding to it, that the company "did not take the request to fix issues seriously enough" given the expectations set before Fable 5 shipped. Multiple sources said the administration had signaled this release would be the first real test of a new executive order on AI guardrails, and that a company positioning itself as the industry's safety leader could not then appear reluctant to address a safety concern without inviting a hard line. One former official warned the controls "will not be easily removed" if Anthropic keeps brushing off concerns raised at the highest levels.

Anthropic disputes that narrative sharply. A source close to the company denied any wellness-retreat unavailability of CEO Dario Amodei, said executives reached the White House within 15 minutes, and insisted there was never a refusal to fix anything. The competing accounts have not been reconciled, which is itself part of the problem: with the underlying letter sealed and the technical paper held privately, the public is left weighing anonymous characterizations against one another.

What to watch

Anthropic's senior technical staff have been in Washington meeting with White House officials, with sources describing daily virtual meetings since the initial outreach. The near-term question is whether a patch, or a face-saving agreement, gets the models switched back on, and how quickly. The larger question is what the episode signals to the rest of the industry and to foreign buyers. Hendrix warned the move is "likely to raise alarms in foreign capitals about the reliability of American AI for critical applications," precisely the export market U.S. policy claims to want to win. If export controls can be invoked overnight against a domestic lab over a contested guardrail, watch for whether other governments and enterprises start treating American AI as a politically conditional product, and whether Congress or the courts demand the kind of transparency this order conspicuously lacked.