Anthropic's Safety Superpower

Anthropic's released Fable and Mythos models have had access suspended for all foreign users due to an export control directive under the US government's national security authority.

The government based this action on the grounds that these models were exposed to jailbreaking techniques capable of identifying and exploiting cybersecurity vulnerabilities. Anthropic has confirmed cases of vulnerability discovery through certain techniques, but claims this is a level that other public models can also perform.

Anthropic's executives are currently in Washington D.C., negotiating to resolve the issue by explaining that this action stems from a misunderstanding. This incident is expected to be recorded as a major case illustrating the conflict between models' powerful capabilities and the resulting Safety and regulation concerns.