According to The Wired on June 16, citing analysis by Tarah Wheeler, security chief at TPO Group, and Harvard University researcher Bruce Schneier, although the Trump administration forced Anthropic to take down Claude Fable 5 and Mythos 5 over national security risks over the weekend, the crackdown cannot stop global spread within 24 months.
Trump administration forced the takedown of Fable 5 and Mythos 5 for “jailbreak” reasons; the relaunch date is undetermined
The report said that Claude Fable 5 was made public to the general public on June 9, but it had specific blocks on issues such as biology and cybersecurity, and its safeguards were deemed capable of being bypassed (jailbroken), triggering the Trump administration’s national security determination. Claude Mythos 5 used a stricter distribution method, initially releasing it privately only to a selected alliance within the “Project Glasswing” working group.
When Anthropic first released it in April under the name “Mythos Preview,” it also warned at the same time about the dual-use nature: in the hands of cybersecurity researchers it can be a defensive tool, while in the hands of malicious actors it could be an offensive weapon. As of the time of the report, Anthropic was still negotiating with the White House, and it is unclear when it will be brought back online.
Anthropic frontline red team lead Logan Graham’s warning statement to WIRED
Logan Graham told WIRED as early as April 2026, when Mythos Preview was released: “The real message is that this is not about this model or Anthropic. We have to prepare now for a world where, over the next 6, 12, 24 months, these capabilities will be widely available.” Graham is Anthropic’s internal lead for its frontline red team, representing the perspective from inside the company.
Tarah Wheeler and Bruce Schneier explain the limits of regulation to WIRED
Tarah Wheeler (TPO Group security chief) told WIRED: “To think no other competitor will develop similar capabilities to Mythos is extremely short-sighted. Other companies are closely following behind Anthropic, and they likely have these capabilities too, but they are holding back, waiting to see how Anthropic will be treated under the current regulatory environment.”
Bruce Schneier (Harvard University researcher) told WIRED: “Smaller, cheaper, open-source models, sometimes on their own and sometimes in collaboration, can match Mythos’s performance with more sophisticated prompting. We should expect other models to match Mythos/Fable’s creativity and resilience within months, with open-source models taking a bit longer.”
Frequently Asked Questions
What are the specific reasons Claude Fable 5 and Mythos 5 were taken down?
According to the report, the Trump administration forced the takedown on the grounds of national security risk. The core reason is that Fable 5’s safeguards can be解除 (jailbroken). Mythos 5 was also taken down at the same time due to its dual-use cybersecurity characteristics (usable for both offense and defense). As of the time of the report, Anthropic is still negotiating with the White House over the conditions for relaunch, and the specific timing is not set.
Why do cybersecurity experts think taking down these two models can only treat the symptoms?
The analyses by Logan Graham (Anthropic), Tarah Wheeler (TPO Group), and Bruce Schneier (Harvard University) all point out that AI capabilities at the same level will, within 6 to 24 months, be widely distributed from multiple sources. This includes commercial competitors who are watching how regulation is handled (Wheeler) and cheaper open-source models (Schneier). Therefore, regulating a single company or a single batch of models cannot solve the structural problem of capability diffusion.
How do the release methods of Mythos 5 and Fable 5 differ?
Fable 5 was made available to the general public (but had blocks on biology and cybersecurity issues). Mythos 5 used a tightly restricted distribution method, initially opening it only to a selected alliance within the Project Glasswing working group. The difference between the two lies in their target users and access thresholds, but both were taken down by the Trump administration due to the issue that their safeguards could be bypassed.
