[Disaster] US Government Forces Anthropic to Kill Fable 5 and Mythos 5 Access Worldwide

[Golden Tara]

Game-change Massive news dropped late on 12 June. The US government issued an export control directive ordering Anthropic to immediately suspend all access to Fable 5 and Mythos 5 for every foreign national, whether inside or outside the United States, including Anthropic's own foreign national employees. The practical effect is a total shutdown for all customers worldwide. Anthropic received the letter at 5:21pm ET and had to move fast. The government cited national security but gave no specific details, only claiming awareness of a jailbreak method. Anthropic reviewed the demonstrated technique and says it amounts to asking the model to read a specific codebase and fix software flaws, which is something other publicly available models including OpenAI's GPT-5.5 already do routinely.

Anthropics position is that this jailbreak is narrow and non-universal and that their defence-in-depth strategy was always the plan, since no model provider can achieve perfect jailbreak resistance. They spent weeks red-teaming Fable with the US government, UK AISI, and private organisations before launch. They even imposed 30-day data retention on Fable customers specifically to catch and shut down attacks. None of that appears to have mattered. Anthropic is complying but is pushing back publicly, arguing that applying this standard across the industry would effectively halt all frontier AI deployments. All other Anthropic models remain available.

Statement on the US government directive to suspend access to Fable 5 and Mythos 5www.anthropic.com

[Di46]

So the government finds a narrow jailbreak that GPT-5.5 can also do, and the response is to shut down a model deployed to hundreds of millions of people. That is not proportionate

[ShawnMichaels_99]

The timing is the thing that gets me. They served the order at 5:21pm with no prior dialogue and demanded immediate shutdown. That is not how you treat a company that voluntarily spent weeks helping you red-team their own product

[ECWAlex98]

I actually think this has more to do with the ongoing Anthropic vs DoD dispute that has been simmering since January. This feels political, not technical

[SlayedRebellion]

Regardless of the politics, Anthropic complied. They did not fight the order or seek an emergency injunction. That tells you they know the legal ground here is shaky for them even if they disagree

[Cobalt Warren]

The Euractiv angle on this is interesting too. Framing it as blocking non-citizens reveals how this export control logic works in practice. It ends up a global shutdown because you cannot selectively block foreign nationals in a cloud product

[Kev5]

Dario's public statement about needing statutory AI oversight with transparent processes is clearly aimed at turning this into a policy debate rather than just a product dispute. Smart move

[Cass93]

Anyone actually relying on Fable for production work right now is in a bad position. There must be enterprise customers absolutely furious about this

[Always_Myles26]

The defence-in-depth strategy they describe is sensible engineering. You assume you will be jailbroken eventually, you monitor, you patch. The government apparently wanted a guarantee of perfect jailbreak resistance that no model can provide

[FreeKickKing11]

I keep wondering what the actual classified concern is. The public rationale is clearly not the whole story. Something else must have been in that letter

[NoLimitsBen18]

If this standard was applied consistently every frontier model currently deployed would need to be pulled. ChatGPT, Gemini, all of them have non-universal jailbreaks. They are just not being targeted right now

[Nadir Compass]

Gov interference just sent AI stocks tumbling unless this unwinds before Monday