US Government Restricts Anthropic's Fable 5 and Mythos 5

The Commerce Department sent a directive placing export restrictions on Anthropic's most powerful models, Fable 5 and Mythos 5, the Wall Street Journal and Bloomberg reported. Anthropic published its own statement on June 12, confirming the directive arrived at 5:21 PM ET from Commerce Secretary Howard Lutnick, citing national security authorities. Reuters and CNBC reported that, to comply, Anthropic disabled access to both models for all customers worldwide -- because U.S. export controls apply to any foreign national regardless of location, Anthropic said it had no practical alternative. Anthropic CEO Dario Amodei and the company's statement addressed Lutnick directly. Fortune reported Amazon CEO Andy Jassy had a phone call with the White House relaying the security concern prior to the directive being issued.

Fortune reported, citing a detailed blog post by security researcher Katie Moussouris (Luta Security), whom Anthropic had hired to review the Amazon research, that the technique triggering the ban was simple: researchers at Amazon asked Fable to "fix this code" for software with known vulnerabilities. When asked to "review the code for security issues" the model refused, but when asked to "fix this code" it produced patches -- and because finding vulnerabilities is necessary to generate fixes, the same process could in principle help an attacker identify flaws. Anthropic's own statement said it reviewed the specific demonstration and found the level of capability was "widely available from other models (including OpenAI's GPT-5.5)" and is "used every day by the defenders who keep systems safe." Anthropic described the jailbreak as narrow and non-universal, adding: "We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result." Moussouris wrote that the vulnerability "cannot meaningfully be fixed, and any attempt would only weaken the model for defense."

An open letter at freefable.org, co-addressed to Commerce Secretary Lutnick and National Cyber Director Sean Cairncross, drew about 100 signatories as of June 15, per Fortune, including executives from Nvidia, Adobe, Zoom, Google, Anaplan, and Sophos. The letter was organized by Alex Stamos (chief security officer at Corridor and former Facebook CSO), with Katie Moussouris among the prominent signatories. TechCrunch quoted the letter: "To pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous." The letter noted that Mythos-class models are "quite good at finding flaws and weaponizing exploits" but "not uniquely good at these tasks," and cited OpenAI's GPT-5.5, Claude Opus and Sonnet, and Moonshot AI's Kimi 2.7 as performing similar code reviews. Axios reported an unnamed source familiar with the administration's thinking said White House tensions were partly inflamed by Anthropic engaging Moussouris -- described by the source as a "radical Democrat" -- and by former CISA director Chris Krebs publicly vouching for her analysis. CNBC reported senior Anthropic staffers met with administration officials to try to resolve the dispute.

Fortune noted Mythos was the first model to successfully complete both cybersecurity test ranges used by the UK AI Security Institute to evaluate hacking capabilities, and was notable for being able to autonomously chain multiple vulnerabilities to potentially orchestrate attacks. Anthropic stated Fable 5 was built with a defense-in-depth strategy and required 30-day data retention for customer interactions to enable jailbreak monitoring. Anthropic's statement said: "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers." The Politico reported in April that federal agencies were quietly evaluating Mythos despite the directive -- an indicator of ongoing tension between government security concerns and AI procurement needs.

This appears to be the first time the U.S. government has retroactively placed export controls on a commercially deployed AI model, according to public reporting. Government use of export restrictions to limit access to a commercial AI model is an escalation beyond prior regulatory moves, including a Department of Defense supply-chain designation reported by CNBC earlier in 2026. The episode highlights a structural tension: defenders argue restricting access reduces their ability to find and fix vulnerabilities, while officials frame the move as a national-security precaution against adversary access.

Track whether the Commerce Department publishes a formal risk assessment tied to the directive; Anthropic's statement noted the letter did not include specific public details. Monitor the outcome of meetings between Anthropic and administration officials per CNBC, and whether the open-letter coalition at freefable.org secures a formal review process or exemptions. Also watch for competitive model releases: the open letter and Anthropic's statement both cite existing models with similar capabilities, a point central to the dispute over whether Fable provides unique uplift.

Reporting to date attributes the export restriction and company action to named news outlets and Anthropic's own statement. Amazon has not issued a public statement on its researchers' role in discovering the jailbreak or Jassy's White House communications. The specific full scope of the jailbreak method beyond the "fix this code" technique described by Moussouris has not been publicly disclosed.