White House's export limits on Anthropic linked to concerns about Chinese access
5 hours ago
- #AI Security
- #Anthropic
- #Export Controls
- The White House imposed export controls on Anthropic's Mythos AI model due to suspicions of access by a China-linked group.
- Anthropic restricted Mythos access to US citizens, leading to the model's complete removal from the market.
- Mythos was designed to find bugs in computer code, posing a risk if exploited by malicious actors.
- Concerns included potential Chinese reverse-engineering (distillation) of the model, threatening US national security.
- An adviser revealed that Amazon alerted the government about a jailbreak vulnerability in Fable 5 (Mythos's consumer version).
- Anthropic's CEO downplayed the jailbreak risk, refusing to fix it, prompting the administration's reluctant export controls.
- Anthropic denies the White House discussed Chinese access to Mythos and prohibits product access from China.
- This incident follows prior conflicts between Anthropic and the White House over AI regulation and autonomous weapons.