On a Friday afternoon in June 2026, the U.S. Commerce Department sent a letter to Anthropic that would send shockwaves through the artificial intelligence industry. The letter invoked an obscure export control directive that effectively banned non-Americans—including Anthropic's own employees—from accessing the company's latest models, Fable 5 and Mythos 5, citing an unspecified national security concern. Anthropic, taken by surprise, swiftly shut down both models for all customers to comply. The result was a stunning demonstration of unilateral government power: no court order, no public hearing, just a letter and immediate compliance.
At first glance, the rationale seemed to center on a security vulnerability—a jailbreak that allowed the models to bypass their guardrails. But as details emerged over the following days, it became clear that the ban was never really about an AI jailbreak. Instead, it appears to have been a retaliatory strike by the Trump administration against a company it had growing tensions with.
The alleged jailbreak
According to a blog post by cybersecurity veteran Katie Moussouris, founder of Luta Security, Anthropic had shared with her a private paper written by security researchers at Amazon. The paper described a way to bypass Fable 5's guardrails by phrasing a query slightly differently: instead of asking the model to "fix this code," researchers asked it to "review code for security issues." The end result was functionally identical—the model provided code fixes—but the second phrasing skirted the safety filters.
Moussouris, a respected figure in the cybersecurity field, was blunt in her assessment: "The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense." She argued that the bypass was trivial and should never have triggered an export control. The distinction between asking for a review versus asking for a fix is one of semantics, not substance. In practice, both are legitimate uses of AI for defensive cybersecurity, and restricting them harms network defenders more than attackers.
Over the weekend, dozens of top security researchers and experts joined Moussouris in calling on the Trump administration to revoke the export control order. They warned that pulling advanced cybersecurity capabilities from U.S. network defenders was "dangerous" and could leave critical infrastructure more vulnerable to attacks.
Political undertones
Citing sources, Axios reported a tense relationship between Anthropic and the Trump administration, marked by "personality differences." The export directive, according to those sources, was driven more by this fractious dynamic than by any technical flaw in the AI products. Justin Hendrix, editor of Tech Policy Press, noted that the move "is likely to raise alarms in foreign capitals about the reliability of American AI for critical applications." The message is clear: AI companies in the United States cannot be trusted to operate without interference from the U.S. government.
The Trump administration has not officially explained why it invoked the export control directive. Was it a misreading of the research paper? Did Amazon CEO Andy Jassy whisper something into senior officials' ears? Or was it a calculated move to pressure Anthropic into submission? The lack of transparency fuels speculation that the White House may not have fully understood the consequences of its own order and is now scrambling to undo the damage.
Historical precedent
Export controls on cybersecurity tools have a fraught history in the United States. During the 2010s, the U.S. government attempted to fix export laws covering tools that could be used for both defense and attack. The language was so broad that it nearly outlawed legitimate security research, requiring years of adjustments to restore balance. The current directive against Anthropic echoes that overreach, but with an added political edge.
Past administrations, both Democratic and Republican, have wielded export controls cautiously, usually targeting foreign adversaries rather than domestic companies. The Trump administration's willingness to target a U.S. AI lab marks a significant escalation. It suggests that the government intends to use its authority not just to protect national security, but also to settle scores and enforce compliance with unofficial policies.
Implications for the AI industry
The ban on Anthropic's models has immediate and long-term consequences. In the short term, customers—many of whom rely on Fable 5 and Mythos 5 for critical tasks—are left scrambling for alternatives. The models' shutdown may also erode trust in American AI, as foreign companies and governments question whether their access could be cut off at any moment due to a political dispute.
Longer term, the precedent is alarming. If a government can force an AI company to take its models offline without due process, then every AI lab in the country is vulnerable to similar pressure. The decision creates a chilling effect on innovation, as companies may hesitate to release powerful models for fear of government reprisal. It also raises questions about the balance of power between the executive branch and independent regulators, as well as the role of the courts in protecting tech companies from executive overreach.
Some industry observers have drawn parallels to the early days of the internet, when the government attempted to restrict encryption software as a munition. That battle ultimately failed, but only after years of legal wrangling and public debate. Here, the government acted swiftly and without public scrutiny, leaving little room for challenge. Anthropic has not indicated whether it will sue, and the administration has offered no path to rescind the directive.
Security researchers have also pointed out that the alleged jailbreak was not a vulnerability at all, but a feature of the model's flexibility. If an AI can be asked to review code, and in doing so provides a fix, that behavior is not a bug—it is exactly what many users want. Restricting that capability weakens the model's utility for defensive security, and arguably makes the nation more vulnerable to cyberattacks. As one expert put it: "This is like banning a fire extinguisher because someone might use it to start a fire."
The incident has also strained relationships between AI companies and the Trump administration, which has been skeptical of some tech giants while cozying up to others. Anthropic, known for its safety-first approach, may have been an easy target for a White House looking to assert dominance over the AI sector. Meanwhile, companies seen as more aligned with the administration may face less scrutiny, creating an uneven playing field.
In the days since the ban, the security community has circulated petitions, op-eds, and technical analyses calling for the order to be revoked. But so far, the administration has remained silent. Whether the pressure will force a reversal, or whether this is the beginning of a broader crackdown on American AI, remains to be seen.
Source: TechCrunch News