China-linked group may have accessed Anthropic's Mythos AI, report says

According to a Semafor report, the White House's export restrictions on Anthropic's advanced AI models—Mythos 5 and Fable 5—were influenced by fears that a China-linked group may have accessed Mythos. The White House has not confirmed the report, and a post by Trump advisor David Sacks focused instead on jailbreaking claims, not China. Anthropic declined to comment but a spokesperson told Semafor that the government didn't raise China during export discussions. If the Chinese government did access Mythos, it could pose serious national security risks, including potential reverse engineering through distillation, where a student model learns from the advanced one.

This is not the first security incident involving Mythos. Anthropic previously stated the model was too dangerous for public release, yet a Discord group managed to access it for two weeks before Anthropic discovered and terminated the breach. The new allegations add to concerns about the security of frontier AI systems and the geopolitical implications of such powerful technology falling into the wrong hands. The incident underscores the ongoing tension between AI safety, export controls, and international competition in advanced AI capabilities.