The US government has ordered Anthropic to pull Claude Fable 5 and Mythos 5, arguing that the models pose a national security risk. The key concern is jailbreakability: according to the White House, a public-facing model like Fable 5 becomes dangerous if its safety guardrails can be bypassed.
Anthropic is still in talks with the White House over the conditions for restoring access, and there is no timeline yet for a relaunch. The order targets one company and two models. Security researchers interviewed in connection with the case argue that the underlying capabilities will not stay contained for long.
Anthropic had already described Mythos as a dual-use system
Mythos 5 was not presented as an ordinary language model. When Anthropic first introduced it in April under the name “Mythos Preview,” the company also warned about its risks. In the hands of security researchers, it could serve defensive work; in the hands of malicious actors, it could become an offensive tool. That distinction sits at the center of the debate.
For that reason, Mythos 5 was initially released only to a selected alliance inside the Project Glasswing working group. Fable 5 was made available to the public, but Anthropic said it had specific blocks around biology- and cybersecurity-related prompts.
Researchers say the real issue is how fast the capability spreads
Back in April, Anthropic frontier red team lead Logan Graham told WIRED that the message was not really about one model or one company. In his view, people should prepare for a world where these capabilities become broadly accessible within 6, 12, and 24 months.
Tarah Wheeler, chief security officer at TPO Group, made a similar point in sharper terms. She said it is extremely shortsighted to assume that no competitor will develop capabilities comparable to Mythos, or to assume that they have not done so already. According to Wheeler, other companies are close behind Anthropic and may already have similar systems, while waiting to see how regulators respond in the current environment.
Smaller, cheaper, and open-source models could catch up
Harvard researcher Bruce Schneier also told WIRED that this reflects a broader technological trend. Smaller and cheaper models, including open-source systems, may at times work alone and at times in combination, with better prompting helping them reach performance comparable to Mythos.
His expectation is that other models could match the creativity and resilience of Mythos and Fable within months, with open-source models following somewhat later. That leaves regulators trying to restrict a product release while the broader capability keeps moving into wider circulation.