Seventy-six of the world’s most prominent cybersecurity experts have issued a blunt warning to the White House: by restricting access to Anthropic’s Fable and Mythos models, the government is effectively disarming the good guys.

The protest follows a surprise order from the U.S. government last Friday, which forced Anthropic to halt the export of its most potent AI models. The directive, issued without public explanation, cited national security concerns. Anthropic complied immediately, pulling the plug on access for users worldwide. Now, the industry is pushing back.

The 'Defensive' Gap

For security professionals, the timing could not be worse. The letter, signed by industry veterans including former Facebook security chief Alex Stamos and cryptographer Jon Callas, argues that these models are not just tools—they are essential infrastructure for finding and patching vulnerabilities before malicious actors do.

"To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous," the letter states. It is a stark assessment. The experts argue that the government has misidentified a standard defensive workflow as a security threat.

The Amazon Paper Controversy

At the heart of the dispute is a non-public research paper from Amazon. The government reportedly relied on this paper to justify the export controls, fearing that the models could be "jailbroken" to perform malicious tasks. However, security researchers who have reviewed the document tell a different story.

Katie Moussouris, a signatory of the letter and founder of Luta Security, argues that the paper does not demonstrate a dangerous bypass. Instead, it shows the model performing its core function: identifying and fixing known vulnerabilities in code.

"Defenders need to be able to ask AI to fix the bugs in a file," Moussouris wrote in a recent blog post. "That is not a guardrail bypass. It is the most valuable thing an AI model can do for defensive security."

A Futile Restriction?

Perhaps the most damning argument in the letter is the claim that the ban is functionally useless. The signatories contend that the same vulnerability-finding capabilities described in the Amazon paper are already present in a wide array of other models, including OpenAI’s GPT-5.5, Anthropic’s own Claude Opus 4.8, and even Chinese-developed models like Kimi 2.7.

If the goal was to prevent the proliferation of these techniques, the experts argue, the government has failed. By restricting only Anthropic’s specific models, the policy creates a vacuum that other, less-regulated models are already filling. It is a policy of optics, not security.

Key Takeaways

  • The U.S. government has banned the export of Anthropic’s Fable and Mythos models, citing national security risks.
  • A group of 76 cybersecurity experts argues the ban prevents defenders from using AI to patch critical software vulnerabilities.
  • Researchers claim the "jailbreak" cited by the government is actually a standard defensive "find and fix" loop that is already possible with other global AI models.

What Happens Next

The industry is now waiting for a response from the White House. The signatories are calling for a transparent, science-based rulemaking process that accounts for the realities of modern software development. With the next round of federal AI policy reviews scheduled for early next month, the pressure is on regulators to prove that these restrictions are based on technical necessity rather than a misunderstanding of how security professionals actually work. Until then, the tools that defenders rely on remain locked behind a government-mandated wall.