Anthropic's Mythos may have suffered an 'embarrassing' leak that made the government 'ban' it

A China-linked group may have gained access to Anthropic's powerful Mythos AI model, according to a report by Semafor. The reported access is said to be one of the factors that may have influenced the White House's decision to impose export restrictions on Anthropic's latest AI systems, Mythos 5 and Fable 5.

While the US government has not officially confirmed the claim, the report has raised fresh questions about the security of advanced AI models and whether potential leaks played a role in the restrictions that effectively cut off foreign access to the technology.

China-linked access concerns reportedly alarmed officials

According to the report, the White House was concerned that a group linked to China may have accessed Mythos. It said such access could pose a national security risk because advanced AI models can be used for cybersecurity research and vulnerability discovery.

Officials were also reportedly concerned that access to the model could allow attempts to replicate its capabilities through AI distillation, a process in which another model learns from a more advanced system.However, neither the White House nor Anthropic has publicly confirmed that China-related concerns directly triggered the restrictions.Anthropic has denied claims that its latest models pose the risks described by some government officials.

A company spokesperson also reportedly told Semafor that China was not mentioned during discussions with the government about export controls.

Previous Mythos security incident comes under spotlight

The latest report has also renewed attention on an earlier security issue involving Mythos. According to The Verge, a Discord group reportedly gained access to the model for about two weeks before Anthropic identified the issue and removed access.The incident was seen as particularly embarrassing because Anthropic has repeatedly described Mythos as a highly capable model that requires strict safeguards and limited access.

The US government recently imposed export restrictions on Anthropic's Mythos 5 and Fable 5 models, preventing foreign governments, companies and individuals from using the technology. The restrictions were introduced amid growing concerns about AI security, cybersecurity risks and the potential misuse of advanced models.