For makers and artists relying on cutting-edge AI tools to audit their own code or secure their creative platforms, the sudden shutdown of access to Anthropic’s most advanced models represents a significant blow. The latest regulatory clampdown has effectively removed the sharpest security scanners from the hands of defenders, forcing a halt to critical vulnerability detection and patching workflows across the industry.
Experts rally against the export ban
Dozens of prominent cybersecurity veterans have signed an open letter demanding the immediate reversal of a new U.S. government export control order. The signatories argue that restricting access to Anthropic’s Fable and Mythos models without a transparent justification is reckless, especially while global adversaries continue to advance their capabilities.
The letter states that this action has stripped defenders of their most potent weapons, leaving them unable to utilise these models to identify flaws or harden their software products. As one signatory noted, removing such high-level capabilities from protectors while threats escalate is a dangerous move.
The regulatory trigger and the immediate fallout
Last Friday, the U.S. administration ordered Anthropic to curtail the export of the Fable and Mythos models, citing national security concerns but offering no specific details on the rationale. In response, Anthropic immediately suspended access to these models for all users worldwide.
Currently, the petition is backed by 76 industry leaders, including Alex Stamos, former chief of security at Facebook; Casey Ellis, founder of Bugcrowd; Jon Callas, a renowned cryptographer and ex-Apple security architect; Paul Vixie, computer scientist; Dino Dai Zovi, former head of applied security engineering at Block; Katie Moussouris, founder of Luta Security; and Rachel Tobac, CEO of SocialProof Security.
Why the models were restricted in the first place
When Mythos was first previewed in April, Anthropic warned that its ability to find security vulnerabilities was so potent that it posed a risk if used by malicious actors or foreign adversaries. Consequently, initial access was limited to approximately 50 companies, a number recently expanded to around 150 organisations across 15 nations.
More recently, Anthropic released Fable, a public iteration of Mythos designed with strict guardrails to prevent use in biology, chemistry, and cybersecurity, as well as to stop attempts to distil the model. These restrictions proved so severe that many experts found the system blocked almost any prompt related to security analysis.
Anthropic suggests the White House’s decision may have stemmed from a report indicating a method to bypass these guardrails and unlock the full power of the Mythos-level capabilities.
Technical reality: A false alarm on jailbreaking
Katie Moussouris, one of the letter’s signatories, reviewed an internal Amazon paper that allegedly demonstrated this bypass. However, she clarified in a blog post that the document did not actually prove a successful jailbreak.
Instead, the researchers simply instructed Fable to repair open-source code containing publicly known vulnerabilities alongside deliberately planted ones, after the model initially refused to review the code for security issues. Moussouris argued that the behaviour described in the paper is inherent to the task and cannot be meaningfully fixed without weakening the model’s defensive utility.
“Defenders need to be able to ask AI to fix the bugs in a file, explain why the fix matters, and write tests that confirm the patch works. That is not a guardrail bypass. It is the most valuable thing an AI model can do for defensive security: executing the find, fix, and test loop defenders run every day.”
This critique was echoed in the open letter, which contends that the method cited in the Amazon paper could likely be replicated on OpenAI’s GPT-5.5, Anthropic’s own publicly available Claude Opus 4.8 and Sonnet models, and even Chinese models such as Kimi 2.7.
A call for science-based regulation
The petitioners are urging for regulations that are enforced transparently and fairly, created through a democratic rule-making process. They insist these rules must be grounded in scientific research conducted by industry and academic experts and applied only to the minimal extent necessary to ensure the safety of the American public.
Key takeaways
- Over 70 cybersecurity experts have signed a formal protest against the U.S. ban on Anthropic’s Fable and Mythos models, arguing it hampers defensive security efforts.
- Claims that a jailbreak method forced the ban are being disputed by industry leaders, who argue the cited research demonstrated a limitation rather than a security vulnerability.
- Signatories warn that similar bypass techniques could apply to other leading models, including GPT-5.5 and Kimi 2.7, suggesting the current restrictions are overly broad.
- The group is calling for future regulations to be based on transparent, science-backed processes rather than vague national security assertions.
