The US government moved in mid-June 2026 to restrict two of Anthropic’s newest AI models — Claude Fable 5 and Mythos 5 — citing national security concerns. The Trump administration issued an export-control directive barring “any foreign national” from using the services, forcing Anthropic to take both models offline. As of publication, the company remains in talks with the White House and has not reached an agreement to restore access.
Anthropic had released Mythos 5 privately to a select working group called Project Glasswing, while Claude Fable 5 — described as a Mythos-grade model — was made available to the general public with built-in restrictions on cybersecurity and biology-related responses. The administration’s concern is that Fable 5’s guardrails can be disabled, effectively unlocking the full capabilities of Mythos 5.
Those capabilities are significant. Since Mythos debuted in April 2026, Anthropic has acknowledged the model can both identify software vulnerabilities and develop methods to exploit them. The company described the tension in a blog post: “The same queries that are beneficial in the hands of cybersecurity professionals and biology researchers could be dangerous if available to malicious actors.”
Experts argue, however, that restricting Anthropic’s models does little to address the broader trajectory. “It’s myopic in the extreme to think that no other competitors to Anthropic will develop similar capabilities,” said Tarah Wheeler, chief security officer at TPO Group. Bruce Schneier, a researcher at Harvard University and the University of Toronto, added that smaller and open-source models can already approximate Mythos-level performance with sophisticated prompting, and that comparable models from other developers are likely months away. OpenAI also conducted a private release of a cybersecurity-focused model in mid-April 2026.
A group of cybersecurity leaders made a similar case in an open letter to the administration on Sunday, calling the export-control directive misguided. “The question is whether a specific restriction meaningfully reduces that risk or whether it mainly slows down the people trying to make systems safer,” said Chris Wysopal, cofounder of Veracode.
Experts are calling on governments to develop broader, more transparent policies for managing advancing AI capabilities in cybersecurity and other sensitive areas, rather than targeting individual models or companies.
Source: WIRED