Cybersecurity expert: AI models as dangerous as Claude Mythos are unstoppable; they will be everywhere within 24 months.

Last week, the Trump administration, citing national security concerns, forced Anthropic to remove two of its most powerful models, Claude Fable 5 and Mythos 5. The reason given was that they "could be jailbroken." However, cybersecurity experts dismissed this, stating that this regulation is only targeting one company and a few models; within 6 to 24 months, similar capabilities will be widely available from multiple sources.
(Background recap: Burning money to grow! An OpenAI leak revealed a "business loss of $20.9 billion," with profits not expected until 2030)
(Additional context: CZ praised Hyperliquid as an amazing invention! But joked that without KYC, it’s too risky: I’ve been through it, they’d better have top-notch lawyers)

Last weekend, the Trump administration, citing national security risks, forcibly removed two of Anthropic’s latest models: Claude Fable 5 and Mythos 5. The White House’s logic was: Fable 5 is publicly accessible, but its safety guardrails (simply put, the model’s mechanisms to refuse harmful questions) can be bypassed, which was deemed a national security risk.

Anthropic is still negotiating with the White House on conditions for re-release, and it’s unclear when they will be back online.

This double-edged sword, Anthropic also knows well

Mythos 5 is not an ordinary language model. Since Anthropic first released it under the name "Mythos Preview" in April, the company has been warning about its risks. Simply put, it’s a defensive tool in the hands of cybersecurity researchers, but could be an attack weapon in the hands of malicious actors.

Because of this, Mythos 5 was initially only privately shared with a select alliance of the Project Glasswing team; while Fable 5 was publicly available, it had specific restrictions on biological and cybersecurity issues.

6 months vs 24 months

Despite the Trump administration’s order to block, Anthropic’s former red team lead Logan Graham made it very clear to WIRED back in April when Mythos Preview was released:

"The real message is, this isn’t about this model or Anthropic. We need to prepare for a world where, within 6, 12, or 24 months, these capabilities will be widely accessible."

Cybersecurity consulting firm TPO Group’s CISO Tarah Wheeler was even more direct:

"Thinking that no other competitors will develop capabilities similar to Mythos, or even that they haven’t yet, is extremely shortsighted. Other companies are closely pursuing Anthropic and likely already possess these capabilities—they’re just waiting and watching to see how Anthropic will be treated under current regulations."

Harvard researcher Bruce Schneier also told WIRED: "This is a broader trend in technology. Smaller, cheaper, open-source models—sometimes alone, sometimes collaborating—can match Mythos’s performance through more sophisticated prompting. We should expect other models to match Mythos/Fable’s creativity and resilience within a few months, with open-source models taking a little longer."