On Friday, a coalition of 76 prominent cybersecurity experts published an open letter protesting the US government’s “dangerous” export control ban on Anthropic’s Fable and Mythos AI models, warning that the sudden restriction strips crucial defensive tools from security researchers worldwide.
Why Security Veterans Are Sounding the Alarm
According to the open letter, the federal intervention actively harms the cybersecurity community by denying defenders access to state-of-the-art AI models. Without these tools, professionals cannot effectively locate software vulnerabilities or fortify digital infrastructure.
“To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous,” the letter warns, highlighting the strategic disadvantage this decision creates.
The unilateral export control order was issued to Anthropic on Friday, citing broad national security concerns. However, the US government failed to provide specific reasons or evidence behind the mandate, according to Anthropic. In immediate response to the directive, the AI safety startup suspended access to both models for all users globally.
The letter has garnered signatures from some of the industry’s most respected figures. Signatories include former Facebook chief of security Alex Stamos; Bugcrowd founder Casey Ellis; renowned cryptographer and former Apple security manager Jon Callas; internet pioneer Paul Vixie; former Block head of applied security engineering Dino Dai Zovi; Luta Security founder Katie Moussouris; and SocialProof Security CEO Rachel Tobac.
The Rise of Mythos and the Fable Guardrails
When Anthropic launched its Mythos model in preview this April, the company itself acknowledged the system’s unprecedented power in identifying security flaws. To prevent malicious hackers or foreign state actors from weaponising the technology, Anthropic heavily restricted its rollout, initially granting access to just 50 companies before slowly expanding the cohort to 150 organisations across 15 countries.
To address wider market demand, Anthropic recently released Fable, a public version of Mythos. Fable was equipped with stringent guardrails designed to block its application in sensitive domains like biology, chemistry, and cybersecurity, as well as to prevent users from reverse-engineering the model. However, these safety measures proved so restrictive that cybersecurity professionals found the model refused to process virtually any security-related prompts.
var playerInstance_jwplayer_6a7a0a6caef62 = jwplayer( “jwplayer_6a7a0a6caef62” );
playerInstance_jwplayer_6a7a0a6caef62.setup({
playlist: “https://cdn.jwplayer.com/v2/media/lv0GaEwB”,
});
Inside the Disputed Amazon ‘Jailbreak’ Paper
Anthropic suggested that the White House’s sudden export ban may have been triggered by an unpublished research paper alleging a method to bypass, or “jailbreak,” Fable’s guardrails to unlock its raw, Mythos-level capabilities.
Katie Moussouris, a key signatory of the protest letter, confirmed she has reviewed the non-public paper, which was authored by researchers at Amazon.
A Flawed Premise for a Global Ban?
However, Moussouris strongly disputes the paper’s findings. Writing in a blog post, she argued that the researchers did not demonstrate a genuine security bypass. Instead, after Fable initially declined to review code for security issues, the researchers simply instructed the AI to repair open-source code containing known and deliberately planted vulnerabilities.
“The behavior described in the paper cannot meaningfully be fixed, and any attempt would only weaken the model for defense,” Moussouris explained. “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.”
The open letter echoes this critique, pointing out that the capabilities highlighted in the Amazon paper are not unique to Fable. The expert group asserts that these same functions can easily be replicated on rival models, including OpenAI’s GPT-5.5, Anthropic’s own public Claude Opus 4.8 and Sonnet models, and even Chinese alternatives like Kimi 2.7.
Speaking to TechCrunch, Moussouris emphasised that other models do not require bypass techniques because they do not feature Fable’s overly restrictive blockages. “The bugs used to demonstrate the techniques in the paper can be found using the other models,” she noted. “Other models that lack the Fable guardrails often won’t refuse the straightforward request to look for security bugs, so they don’t need a bypass.”
A Call for Democratic and Scientific AI Regulation
Rather than arbitrary bans, the coalition of experts is calling for transparent, evidence-based regulations. They urge the US government to establish a democratic rulemaking process grounded in rigorous scientific research conducted by industry and academic specialists. Such regulatory powers, they argue, should only be deployed to the minimum extent necessary to safeguard public safety without crippling defensive cyber capabilities.