US Government Orders Anthropic to Suspend Access to Fable 5 and Mythos 5 AI Models The clash over Anthropic’s most advanced AI models has turned a technical safeguard dispute into a global test of how far governments will go to control frontier systems.

On June 9, Anthropic launched Fable 5, a guarded public version of its powerful Mythos 5 cybersecurity model, touting it as its most capable release to date. Within days, the Trump administration grew alarmed by reports that Amazon security researchers could coax the model into surfacing code vulnerabilities, information they said might aid cyberattacks. On June 12, Commerce Secretary Howard Lutnick sent an export-control directive barring use of Fable 5 and Mythos 5 by any foreign national, including Anthropic’s own staff, forcing the company to disable both models worldwide to comply.

Anthropic publicly disputed the move, saying the government had provided only “verbal evidence of a potential narrow, non-universal jailbreak” and arguing that similar flaws can be exposed by other public models. In its formal statement, the company stressed it had “instituted strong safeguards” and that no tester had found a universal jailbreak, adding that perfect resistance is “not currently possible for any model provider.”

Inside the administration, officials framed the action as a national security necessity and part of a broader effort to treat top-end AI models as strategic assets. Some privately accused Anthropic of “recklessness” in pressing ahead with launch despite warnings, according to accounts of internal criticism. Concerns reportedly extended beyond the jailbreak itself to fears of Chinese access to Mythos and the possibility of reverse‑engineering the model.

Over the ensuing weekend, the standoff escalated. Senior officials held tense calls with CEO Dario Amodei in what one account described as a frantic 24‑hour push to secure a voluntary takedown before resorting to export controls. After the order landed, Anthropic flew senior technical staff to Washington for crisis talks with the Commerce Department, hoping to demonstrate its cybersecurity safeguards and win back permission to operate.

Beyond Washington, the shutdown triggered a wave of backlash and geopolitical soul‑searching. Cybersecurity leaders warned that removing the most advanced defensive tools from “defenders” while adversaries keep building “is not safety, it is sabotage,” in an open letter calling for reversal of the ban. Another group of experts argued that curbing Fable’s ability to generate “proofs of concept” for vulnerabilities would weaken U.S. cyber defenses and deter companies from building robust security tooling into their models.

Allies and competitors seized on the incident to press the case for technological independence. European startup Mistral said the episode underscores the risk of letting “another country” hold the keys to critical AI infrastructure, bolstering its pitch for sovereign, deploy‑anywhere models. In India, where Anthropic had rapidly grown, the sudden cutoff “landed as a warning shot” and fueled calls for a multi‑billion‑dollar sovereign AI fund and greater reliance on open‑source alternatives. Canadian prime minister Mark Carney compared the concentration of AI capabilities in a few foreign systems to pre‑2008 financial “model risk,” urging countries to diversify and build redundancy.

Within the AI industry, the affair is seen as a watershed. Commentators noted that Fable 5 briefly topped benchmarks before being yanked, leaving developers with weaker—but still available—rivals like GPT‑5.5. Others warned that the U.S. has now shown it can pull a leading model “with a swift and unilateral action,” a precedent that could hang over every future deployment.

As Anthropic and U.S. officials continue negotiations, both sides insist they want a resolution. But the outcome will shape not only the fate of Fable 5 and Mythos 5, but also how nations balance access, security, and sovereignty in the next generation of AI.

Continue reading https://foxvector.com/stories/019ed2ba-5f5e-0f18-7080-3a0d59ae08ec