Encryption, spyware, and now Mythos: History shows why cyber export control doesn’t work
TL;DR
The White House ordered Anthropic to restrict Fable and Mythos for users outside the US and for foreign nationals inside the country. Anthropic then pulled both models offline on short notice. The reported triggers were concern over a South Korean telecom partner and Amazon flagging a Fable 5 safeguard issue. Anthropic says this was not a broad jailbreak, but a narrow, already patched problem.
Nauti's Take
Export controls sound decisive, but in cyber they often create delay and paperwork more than real containment. If Mythos is genuinely that powerful, a national border is a weak safety model.
Strong audits, clear misuse thresholds, traceable access controls, and vendor accountability would matter more. Otherwise, the winner is not security, but whoever builds the best workaround structure.
Briefingshow
The case shows why traditional export control is awkward when the controlled thing is software or model capability. Hardware supply chains can be slowed; code, expertise, and substitute models tend to route around borders faster. For AI labs, national security policy is becoming part of product design, not just legal paperwork.
Sources