On Friday, the US Commerce Department forced American AI firm Anthropic to pull its flagship Fable 5 and Mythos 5 models offline globally after issuing a sudden national security export ban, sparking fears of unprecedented political interference in the tech sector.
This unilateral enforcement letter, which effectively blacklisted non-US citizens—including Anthropic’s own staff—from accessing the systems, serves as a stark warning to Silicon Valley about the reach of Washington’s regulatory power.
A sudden shutdown under regulatory pressure
To comply with the unexpected directive, Anthropic shut down both of its premier models for all global users. The Commerce Department invoked an obscure export control mechanism, citing vague national security risks. Anthropic suspects the order relates to an alleged bypass of the model’s safety guardrails, though the government has refused to make the letter public or provide specific details. This swift executive action bypassed judicial oversight entirely, demonstrating how easily the Trump administration can take advanced commercial software offline.
Personality clashes over technical failures
The sudden intervention highlights that the artificial intelligence industry is no longer shielded from aggressive state interference. According to sources cited by Axios, the export directive stemmed from deep-seated “personality differences” between Anthropic executives and the Trump administration, rather than any genuine technical flaw within the AI models themselves. This friction points to a retaliatory motive rather than a measured security response.
The reality behind the ‘jailbreak’ paper
Further details emerging over the weekend undermine the official justification for the ban. Cybersecurity expert Katie Moussouris, founder of Luta Security, revealed in a blog post that Anthropic had shared a private research paper with her. The document, authored by security researchers at Amazon according to the Wall Street Journal, detailed a guardrail bypass in Fable 5.
However, Moussouris argued that the bypass was trivial and did not warrant an export ban. The issue boils down to asking the AI to “review code for security flaws” versus asking it to “fix this code”—different prompts that yield virtually identical results. She warned that the behaviour cannot be easily patched without weakening the model’s defensive capabilities, labelling the government’s response as hasty, heavy-handed, and counterproductive.
Broader implications for cybersecurity and trust
Moussouris and dozens of other prominent security researchers have urged the Trump administration to rescind the order, warning that denying American cyber-defenders access to advanced tools is dangerous.
Historically, US administrations have struggled with technical literacy when drafting policy. In the 2010s, poorly worded export controls designed to restrict cyberweapons nearly outlawed legitimate vulnerability research.
A damaging precedent for US tech
This latest administrative overreach could severely damage international trust in American technology. Justin Hendrix, editor of Tech Policy Press, warned that the move will raise serious alarms in foreign capitals regarding the reliability of US-made AI for critical infrastructure. The underlying message is that American AI firms cannot guarantee stable service free from sudden government interference.
The White House has yet to clarify its motives. Officials may have misunderstood the technical research, acted on complaints from rival tech executives, or used the order to apply political pressure on a partner with whom they have a fractious relationship. Regardless of the intent, the administration has set a highly volatile precedent, showing that any tech firm could find its products pulled offline overnight if they fall out of political favour.