Share

    • US Government Orders Anthropic to Halt Access to Fable and Mythos AI Models

    By Any June 15, 2026
    The U.S. government has ordered Anthropic to suspend access to its most advanced AI models, Claude Mythos 5 and Claude Fable 5, citing national security concerns and the discovery of a "jailbreak" vulnerability. Anthropic has complied with the export control directive, which restricts access for foreign nationals and led to the models being taken offline for all users, though the company has expressed disagreement with the government's action.
    US Government Orders Anthropic to Halt Access to Fable and Mythos AI Models

    US Government Orders Anthropic to Halt Access to Fable and Mythos AI Models Anthropic’s most advanced AI systems, Claude Mythos 5 and Claude Fable 5, went from record‑breaking launch to forced shutdown in three days, exposing a deep rift between fast‑moving AI labs and governments trying to treat cutting‑edge models as national security assets.

    From breakthrough launch to export controls

    Anthropic unveiled Fable 5, a public variant of its Mythos‑class system, on June 9, touting it as its “most powerful cybersecurity model” and quickly pushing it to the top of public benchmarks and the Chatbot Arena leaderboard. Within days, the Trump administration moved to block foreign access to Mythos 5 and Fable 5 after reports that another company had “jailbroken” the model to bypass safety guardrails, triggering fears of automated hacking at scale.

    On June 12, Commerce Secretary Howard Lutnick sent Anthropic an export‑control letter covering any foreign person inside or outside the US, including Anthropic’s own staff. Anthropic responded that “the net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance,” while emphasizing that access to its other models was unaffected.

    Government, Amazon and security fears

    Behind the scenes, Amazon security researchers told the White House they could prompt Fable 5 to produce information useful for cyberattacks, and CEO Andy Jassy relayed those concerns to senior officials. Axios reported that calls from Amazon and several other companies led to sweeping export rules that made Fable 5 unavailable just days after release.

    Officials also worried Mythos could help exploit vulnerabilities in critical infrastructure, including the global financial system, and later reports suggested fears that a China‑linked group might have accessed Mythos and could distill it into their own systems.

    Anthropic and expert pushback

    Anthropic said the government provided no written details of its national security concern and that, based on a demonstration it reviewed, the issue was a “potential narrow, non‑universal jailbreak” exposing only “previously known, minor vulnerabilities” that other public models can also find. It argued that treating such a finding as grounds for recall “would essentially halt all new model deployments for all frontier model providers.”

    The company publicly disagreed with the decision but complied, calling it “a misunderstanding” and warning that if the same standard were applied industry‑wide it would freeze frontier AI progress. Senior technical staff were quickly flown to Washington to try to resolve the dispute with the White House.

    A group of prominent cybersecurity leaders, led by former Facebook CSO Alex Stamos, urged the administration to reverse course, arguing that the ban “has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.” They stressed that Fable’s ability to generate proofs‑of‑concept for exploits helps blue‑team defenders as much as it could aid attackers.

    Global and market repercussions

    Because the controls covered foreign nationals, Anthropic’s second‑largest market, India, lost access overnight, just one day after Tata Consultancy Services announced a major partnership to train 50,000 employees on Claude. The shutdown has intensified India’s debate over “sovereign AI,” with some leaders calling for a multibillion‑dollar national AI fund and a pivot toward open‑source models to avoid dependence on US providers.

    Canadian prime minister Mark Carney compared the episode to 2008‑style systemic financial risk, warning that over‑reliance on a few frontier models creates “model risk” for entire economies and calling for diversification and redundancy in AI infrastructure.

    Developers and companies, meanwhile, are scrambling for alternatives: Fable 5 had briefly “crushed” OpenAI’s GPT‑5.5 on key coding and software‑engineering benchmarks before the recall, leaving GPT‑5.5 as the strongest generally available model by default rather than by improvement.

    AI lab perspective

    In its official statement, Anthropic underscored that extensive red‑teaming with US and UK governments and third‑party testers found no “universal jailbreak,” and that “perfect jailbreak resistance is not currently possible for any model provider.” The company reiterated support for statutory powers to block unsafe deployments, but only via processes that are “transparent, fair, clear and grounded in technical facts” — standards it says were not met in this case.

    As negotiations continue in Washington, the Fable and Mythos shutdown has become a global test case for how states will balance national security, market competition and access to the most capable AI systems.

    Continue reading https://foxvector.com/stories/019ecc48-2396-1853-7185-349ec531763c

    Write a comment