Anthropic Fable 5 Ban Sparks Crisis as Cybersecurity Experts Condemn Government Intervention

The US Commerce Department on Friday invoked an obscure export control directive to force Anthropic to pull its Fable 5 and Mythos 5 AI models offline, citing unspecified national security concerns. The move has triggered a policy crisis and drawn sharp criticism from cybersecurity experts.

The directive banned non-Americans, including Anthropic employees, from accessing the models. Anthropic complied by shutting down both models to all customers. The letter has not been made public, and the company believes it is related to a reported guardrail bypass, but the government has not provided specifics.

Guardrail Bypass Was Not a Jailbreak, Researchers Say

According to cybersecurity veteran Katie Moussouris, who reviewed a private paper from Amazon researchers, the alleged bypass involved prompting the model to "fix this code" rather than "review code for security issues." Moussouris said the behavior cannot meaningfully be fixed and that the export control was hasty and misguided. The Register reported that the bypass was simple and not a jailbreak.

• The directive was invoked under an obscure export control rule.
• Anthropic shut down Fable 5 and Mythos 5 to all customers.
• Roughly 100 cybersecurity experts signed an open letter calling for the ban to be reversed.
• Anthropic senior staff are meeting with Commerce Department officials on Monday.
• The Trump administration has not confirmed the reasoning behind the directive.

Broader Implications for AI Industry and National Security

The incident sets a dangerous precedent for government interference in AI product releases. Justin Hendrix of Tech Policy Press warned it raises alarms about the reliability of American AI for critical applications. The dispute highlights tensions between the Trump administration and Anthropic.

What comes next: The Monday meeting could determine whether the ban is lifted or escalates further. The cybersecurity community argues that pulling advanced AI tools from defenders while adversaries continue building is sabotage. The outcome will test the boundaries of executive power over emerging technology.