Claude Fable 5 And Mythos 5 Just Got Shut Down By The Government!

By In The World of AI

Categories: AI

Summary

The US government shut down Anthropic's Fable 5 and Mythos 5 models citing national security concerns over a jailbreak technique—but Anthropic argues the vulnerability is narrow, already exists in competitors' models like GPT-4.5, and shouldn't trigger full model recalls. This sets a dangerous precedent for regulatory overreach that could freeze all frontier AI development.

Key Takeaways

  1. Government issued export control directive blocking all foreign nationals from accessing Fable 5/Mythos 5, including Anthropic's own foreign employees—a complete shutdown rather than regional restriction or throttling.
  2. The disputed 'jailbreak' is simply asking the model to read a codebase and fix software flaws—a capability already publicly available in OpenAI's GPT-4.5 and used daily by security defenders.
  3. Anthropic employed 'defense in depth' strategy: make jailbreaks either narrow or expensive, pair with heavy monitoring, and enforce 30-day data retention to catch and patch attacks fast—but even they admit perfect jailbreak resistance is impossible.
  4. Anthropic conducted thousands of hours of red-teaming with US government, UK AI Safety Institute, and external organizations before launch—yet models still got leaked and sold on Chinese platforms before official release.
  5. If this regulatory standard applies uniformly across all frontier labs, it would essentially freeze new model releases industry-wide—making this precedent a systemic threat to AI development, not just Anthropic.

Related topics

Transcript Excerpt

So, this one is pretty wild and it's moving as I'm recording. So, keep that in mind. Today, Enthropic put out a statement saying the US government has ordered them to shut off access to Fable 5 and Methos 5, not throttle it, not restricted to certain regions, basically shut it off completely for every customer. And the reason is the part that makes this different from anything we have covered. This came through an expert control directive citing national security. The order blocks access by any foreign national whether they're inside the US or outside it. And that even includes Enthropic's own foreign national employees. Enthropic says the only way to actually comply with this is to pull the models entirely for everybody. Their other models, Opus, Sonnet, Haiku, all that stays up is just F…

More from In The World of AI