Cybersecurity Experts Push Back on U.S. Ban of Anthropic’s Fable 5 Model

The Trump administration levied export controls on Anthropic's Fable 5 AI model on June 12, 2026, citing concerns the model could be jailbroken to assist foreign adversaries with offensive hacking. The Department of Commerce blocked foreign sales of the model after reports from Amazon and a researcher claimed to have bypassed its safety guardrails within days of release.

Dozens of cybersecurity experts signed an open letter on Monday calling on the White House to lift the restrictions. The letter, organized by researcher Katie Moussouris, argues that the demonstrated jailbreak techniques do not unlock unique offensive capabilities that are unavailable in other frontier models like OpenAI’s Daybreak or Chinese models like Kimi 2.7.

Experts say the ban hurts defenders, not attackers

Moussouris, a well known figure in the field who has advised the Wassenaar Arrangement, reviewed third party research on Fable 5's guardrails. The research showed a multistep manual process tricked the model into generating automated scripts for testing software patches. She called this a standard defensive task, not a bypass of safety controls.

“That is the most valuable thing an AI model can do for defensive security: executing the find, fix, and test loop defenders run every day,” Moussouris wrote. She described the export restrictions as “heavy handed” and “misguided” in a post published Monday.

• Anthropic subjected Fable 5 to 1,000 hours of internal and external red team testing, reporting no universal jailbreaks were found.
• The model defaults to older, less powerful models on sensitive topics like cybersecurity and biological warfare.
• OpenAI’s Daybreak model offers similar vulnerability discovery and patching capabilities but was not included in the Commerce Department’s restrictions.

Internal contradictions in the government’s case

The experts’ letter questions whether the reported jailbreak demonstrations would even qualify as offensive capabilities. The researchers noted that similar results can be reproduced in other commercial and open source models. “AI has been finding bugs and generating working exploits at superhuman levels since last year,” the letter states.

Moussouris was provided a copy of third party research on guardrail bypass techniques for Fable 5. The researchers asked three Claude models to review vulnerable open source code. Fable 5 initially refused, but after a multistep process it turned output into automated scripts. The experts argued this is foundational for cybersecurity defense, not an anomalous threat.

Anthropic shut off the models for all users after the Commerce Department’s decision, as the company works to persuade the White House to change course.

The open letter, signed by dozens of practitioners, calls the export controls an unprecedented action. The signatories include cybersecurity professionals who use frontier models daily for defense. They say Fable 5’s guardrails have been notoriously oversensitive, becoming a source of humor in the cyber community at launch when users could not get the model to perform basic defensive tasks.