Anthropic disables Claude Mythos 5 and Fable 5 after US export

Anthropic announced at 5:21 pm ET on June 12, 2026 that both Claude Mythos 5 and Claude Fable 5 are disabled for all customers after a U.S. Commerce Department export‑control directive cited national‑security concerns [2]. The order bars any foreign person, inside or outside the United States, from accessing the models, forcing the company to terminate existing sessions and redirect new requests to older models such as Opus 4.8 [2].

The two models were launched only days earlier. Claude Mythos 5 is the unrestricted, cyber‑offense‑capable version of Anthropic’s latest AI, designed to autonomously discover and exploit software vulnerabilities. Access was limited to about 40 vetted organizations through Project Glasswing, including Microsoft, Nvidia, Cisco, Google Cloud and AWS [1][3]. Claude Fable 5, released on June 9, 2026, stripped out the cyber‑offense toolkit and added safety guardrails, offering the same general reasoning and coding abilities to the public at $10 per million input tokens and $50 per million output tokens [1][3].

The export‑control action was triggered after a separate company reportedly jail‑broke Mythos, prompting Commerce Secretary Howard Lutnick to issue a letter demanding a license for any export, re‑export, or domestic transfer of the models [2]. Anthropic’s statement says the government’s concern centers on a narrow jailbreak that allowed the model to uncover a few known, minor vulnerabilities—capabilities that other public models, including OpenAI’s GPT‑5.5, already possess [2]. The company disputes the severity of the issue but complied to avoid violating the directive [2].

For the crypto ecosystem, the shutdown removes a potential defensive tool that could have helped DeFi protocols identify deep‑layer flaws in key‑management systems, oracles and cross‑chain bridges [1][3]. While the restricted Mythos 5 remains unavailable to the broader market, its existence underscores a looming risk: if such powerful AI were to leak or be misused, the attack surface for blockchain applications could expand dramatically [1]. The episode also highlights the regulatory friction that may shape the rollout of advanced AI security tools across industries.

The real question now is whether Anthropic will secure a license and restore access, or whether the export controls will push the crypto security community to seek alternative AI solutions before the next wave of vulnerability‑research models emerges.