Experts call Fable 5 export controls misguided after jailbreak reports
Cybersecurity experts, including Katie Moussouris, say third-party research has not demonstrated that Anthropic's Fable 5 model provides unique hacking capabilities, pushing back on the Trump administration's decision to impose export controls on the model.
Score breakdown
The expert pushback challenges the factual basis of the Commerce Department's export controls, with Moussouris arguing the research cited by the administration demonstrates defensive security capabilities rather than a genuine bypass of Fable 5's safeguards.
- 01The Trump administration's Department of Commerce imposed export controls on Anthropic's Fable 5 AI model.
- 02The controls were triggered by jailbreak reports from Amazon and a cybersecurity researcher shortly after Fable 5's public release.
- 03Anthropic shut off the models for all users while attempting to convince the White House to reverse the decision.
The Trump administration's Department of Commerce imposed export controls on Anthropic's Fable 5 AI model last Friday, citing alarm over reports from Amazon and a cybersecurity researcher who claimed to have jailbroken the model within days of its public release. The administration's reasoning was that if U.S. researchers could bypass the model's safeguards, foreign adversaries could as well. In response, Anthropic shut off the models for all users while working to persuade the White House to change course. Anthropic had already taken steps to limit risks around Fable 5, including declining to release it publicly, funneling access to organizations for cyber defense, and building guardrails that default answers to older, less powerful models on sensitive topics like cybersecurity and biological warfare.
Cybersecurity expert Katie Moussouris, who previously provided technical expertise to the Wassenaar Agreement, reviewed third-party research on the guardrail bypass techniques provided to her by Anthropic.
Cybersecurity expert Katie Moussouris, who previously provided technical expertise to the Wassenaar Agreement, reviewed third-party research on the guardrail bypass techniques provided to her by Anthropic. That research involved asking three Claude models — Fable 5, Mythos, and Claude Opus — to review batches of known, vulnerable open source code for security issues. While Fable 5 initially refused, researchers used "a multistep and manual process" to get it to produce automated scripts for testing patches. Moussouris argued this does not constitute a guardrail bypass, writing that the ability to "fix the bugs in a file, explain why the fix matters, and write tests that confirm the patch works" is "the most valuable thing an AI model can do for defensive security." She called the export restrictions "heavy handed" and "misguided."
Senator Mark Warner, D-Va., echoed concerns about the administration's process, acknowledging that export restrictions on frontier AI models may sometimes be warranted but arguing they must be "grounded in a transparent, risk-based process with clear rules and consistent standards." He called on Congress to pass a statutory framework for testing and approving frontier AI models based on transparency, predictability, and fairness.
Key facts
- 01The Trump administration's Department of Commerce imposed export controls on Anthropic's Fable 5 AI model.
- 02The controls were triggered by jailbreak reports from Amazon and a cybersecurity researcher shortly after Fable 5's public release.
- 03Anthropic shut off the models for all users while attempting to convince the White House to reverse the decision.
- 04Anthropic had already limited Fable 5's availability, funneling it to cyber defense organizations and building guardrails that default to older models on sensitive topics.
- 05Katie Moussouris reviewed third-party research showing researchers used 'a multistep and manual process' to get Fable 5 to produce patch-testing scripts from vulnerable code.
- 06Moussouris called the demonstrated capabilities standard defensive security workflows, not a guardrail bypass, and labeled the export controls 'heavy handed' and 'misguided.'
- 07Senator Mark Warner called for a statutory framework grounded in transparency, predictability, and fairness for evaluating frontier AI models.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 16, 2026 · 23:11 UTC. How this works →