Cybersecurity Experts Rally Against ‘Risky’ Government Prohibition on Anthropic’s Advanced AI Models

A coalition of numerous cybersecurity professionals, including several prominent figures from the field, has addressed an open letter to the U.S. government, urging it to rescind the export control restrictions on Anthropic’s Fable and Mythos models.

The letter emphasizes that this decision has deprived cybersecurity defenders of the best tools available, hindering their ability to identify vulnerabilities and enhance the security of their software and products.

The letter states, “It is perilous to withdraw top capabilities from defenders without a valid justification, especially when our opponents are swiftly advancing.”

Recently, the U.S. government mandated Anthropic to restrict the export of its Fable and Mythos models, citing national security reasons without providing detailed explanations for the directive, according to the company. Consequently, Anthropic ceased access to these models for all global users.

Currently, the letter has garnered support from 76 cybersecurity experts, including notable names such as Alex Stamos, former security chief at Facebook; Casey Ellis, founder of Bugcrowd; Jon Callas, well-known cryptographer and ex-security design manager at Apple; Paul Vixie, a prominent computer scientist; and others from various cybersecurity backgrounds.

When Mythos was introduced in preview earlier this year, Anthropic claimed it had exceptional capabilities in identifying security vulnerabilities. To prevent adversaries from exploiting this power, the company restricted access, initially granting it to about 50 companies which has now expanded to around 150 organizations across 15 countries.

Last week, Anthropic unveiled Fable, a public iteration of Mythos, asserting it includes rigid safeguards to prevent its use in cybersecurity, as well as in biology and chemistry. However, cybersecurity professionals noted that these safeguards were so stringent that they effectively blocked almost all prompts related to cybersecurity.

Anthropic suggested that the export control order from the White House might have stemmed from a report indicating a vulnerability that could potentially bypass Fable’s restrictions, unlocking the more advanced Mythos capabilities.

Katie Moussouris, one of the experts who signed the letter, indicated that the method for bypassing Fable was illustrated by researchers at Amazon in an unpublished study she has reviewed.

However, she clarified in a blog post that the paper did not provide evidence of a genuine jailbreak. Instead, the researchers prompted Fable to correct open-source code that featured public vulnerabilities alongside “deliberately inserted vulnerabilities,” following the model’s initial refusal to analyze the code for security flaws.

“The issues outlined in the paper cannot be adequately remedied, and any intervention would only compromise the model’s efficacy for defense,” Moussouris asserted. “Defenders must have the ability to direct AI to correct flaws in code, justify the importance of these corrections, and create tests to ensure that patches are effective. This is not a bypass of safeguards; it represents the most beneficial function an AI model can provide for defensive security—facilitating the find, fix, and test cycle that defenders engage in daily.”

Moussouris’ observations were further supported by the open letter, stating that the technique described in the Amazon paper “could be replicated” using OpenAI’s GPT-5.5, Anthropic’s publicly accessible Claude Opus 4.8 and Sonnet, as well as even Chinese models like Kimi 2.7.

The letter also urged for transparent and fairly applied regulations formulated through “a democratic rule-making process,” grounded in rigorous scientific research by industry and academic experts, and applied only to the minimum extent necessary to safeguard the American populace.