On June 12, Anthropic suspended access to its Claude models Fable 5 and Mythos 5, just three days after release. The move came directly from a US government export control directive, prohibiting use by non-US nationals.
Mythos is Anthropic’s most powerful frontier model, previously restricted due to advanced hacking capabilities. Fable incorporated additional safeguards, but these were quickly compromised.
The escalating conflict between Anthropic and the Trump administration predates this directive. The White House had accused the company of creating "woke AI" and clashed with CEO Dario Amodei over refusing Pentagon access for domestic surveillance and autonomous weapons systems.
The immediate trigger appears to be a jailbreak. Anthropic believes the government acted after Amazon engineers demonstrated a method to bypass Fable's safeguards. Within 48 hours, a researcher further exposed Fable 5’s full system prompt online.
Oxford economist Maximilian Kasy describes modern AI systems as alchemical: successful but not fully understood. This fundamental opacity makes regulation exceptionally difficult. A recent executive order demands developers share models for pre-release review, signaling the administration no longer trusts companies to evaluate their own risks independently.