techcrunch.com web signal

US Forces Anthropic to Pull Fable 5 and Mythos 5

5 sources tracking this story
anthropic dario amodei safety regulation ai-policy frontier-lab export-controls

Key insights

  • An administration official told Axios the government acted after an unnamed company claimed it could jailbreak Mythos 5, citing national security risk.
  • Anthropic received the directive at 5:21pm ET on June 12 and pulled both models within hours, three days after their public launch.
  • The NSA had previously authorized Mythos 5 for classified network use before the administration ordered its global shutdown, a contradiction The Next Web reports.

Why this matters

The US government's June 12 export control directive ordering Anthropic to disable Fable 5 and Mythos 5 globally sets the first precedent for a government-ordered recall of a commercially deployed frontier AI model. Anthropic's own statement disputes the standard applied, arguing the triggering jailbreak was narrow and non-universal and that the same capabilities are present in other publicly deployed models including GPT-5.5. Fortune's reporting frames the action within a documented pattern of Trump administration hostility toward Anthropic, including the Pentagon's March 2026 'supply chain risk' designation, raising questions about whether the recall rests on purely technical security grounds. The Next Web reveals a direct contradiction at the center of the action: the NSA had authorized Mythos 5 for use on classified networks before the same administration ordered its global shutdown.

Summary

Anthropic received a US export control order Friday at 5:21 PM ET, disabling Claude Fable 5 and Mythos 5 globally. Fable 5, rated by Vals AI as the most capable public AI model, had launched just three days prior. The trigger was a jailbreak: prompting the model to read a codebase and identify software flaws. Anthropic said the same capability exists in OpenAI's GPT-5.5. Essentially: (Anthropic, the US government) are now at odds over the risk framing Anthropic marketed. - Project Glasswing, Mythos' restricted program with roughly 50 vetted organizations including Amazon and CrowdStrike, was also shut down. - Anthropic's own disclosures noted Mythos found flaws in every major OS and browser. Sam Altman had called Anthropic's approach 'fear-based marketing' in April 2026. That framing may have given regulators their justification.

Potential risks and opportunities

Risks

  • Project Glasswing participants (Amazon, Apple, Google, Microsoft, CrowdStrike) face an indefinite halt to Mythos-dependent defensive cybersecurity work with no disclosed restoration conditions.
  • If the government's jailbreak-finding standard is upheld, all major AI labs face new precedent for post-deployment recalls without requiring evidence of actual misuse.
  • Fable 5's global user base loses access to what Vals AI rated the most capable public AI model, creating immediate market exposure for Anthropic as users migrate to OpenAI's GPT-5.5.

Opportunities

  • OpenAI is positioned to capture Anthropic enterprise and consumer users during the shutdown window, particularly given GPT-5.5 was cited by Anthropic itself as a comparable-capability alternative.
  • Legal and compliance consultancies specializing in US export control regimes face sharply increased demand as this action establishes the mechanism can target domestic frontier AI deployments.
  • Cybersecurity firms in Project Glasswing, particularly CrowdStrike, could gain outsized influence in shaping the government's framework for acceptable AI security risk given their vetted-organization status.

What we don't know yet

  • The specific US agency that issued the export control order is not confirmed in public reporting, leaving the enforcement chain unclear.
  • No timeline or compliance conditions for restoring access to Fable 5 or Mythos 5 have been publicly disclosed by Anthropic or the government.
  • Whether Project Glasswing participants (Amazon, Apple, Google, Microsoft, CrowdStrike) received advance notice or had Mythos access cut simultaneously with the public is unreported.

What others are reporting

Coverage cluster as of 8h after publish

  1. Anthropic Read →

    First-party technical defense: describes a 'defense in depth' safety architecture and argues the jailbreak was narrow, warning that the applied standard would halt all frontier model deployments industry-wide.

    We are complying with the government's legal directive and are removing access to Fable 5 and Mythos 5 for all users.
  2. Fortune Read →

    Situates the recall inside a broader pattern of Trump administration antagonism toward Anthropic, including the Pentagon supply chain risk designation; features former officials questioning the policy's internal consistency.

    We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.
  3. The Next Web Read →

    Reports the central contradiction: the NSA authorized Mythos 5 on classified networks before the same administration ordered its global shutdown; also documents the three-day launch-to-recall window.

    essentially halt all new model deployments for all frontier model providers
  4. Heise Online Read →

    Adds international regulatory contrast: EU and German frameworks use market surveillance and documentation obligations rather than single-model shutdowns; notes the US authority disclosed no specific technical details.

    The authority did not provide specific technical details regarding the stated national security concerns.

Originally reported by techcrunch.com

Read the original article →

Original headline: US Government Forces Global Shutdown of Anthropic Fable 5 and Mythos 5 in First-Ever Frontier AI Recall