Anthropic released Fable 5 with safety guardrails, but after a jailbreak method was found, the US government issued an export control directive citing national security, blocking all customer access. Anthropic argues the discovered vulnerabilities are minor and also found in other models.
Anthropic released Fable 5 with safety guardrails, but after a jailbreak method was found, the US government issued an export control directive citing national security, blocking all customer access. Anthropic argues the discovered vulnerabilities are minor and also found in other models.
Anthropic had previously restricted the release of Mythos Preview, citing its dangerous capabilities. Fable was released as a safer version of Mythos, but a jailbreak was discovered shortly after release. The government ordered immediate access suspension due to national security concerns, and Anthropic is disputing the order while seeking resolution in Washington.
This incident highlights the tension between AI model safety and government regulation. The conflict between Anthropic and the US government was anticipated, and this case could set an important precedent for future AI regulation.
HN comments noted that Anthropic's Mythos model outperforms existing open-source models in detecting security vulnerabilities, but the difference is not as absolute as the company claims. Some commenters argued that Mythos' performance actually stems from post-processing that discovers and chains vulnerabilities, and that open-source models could find similar weaknesses. Support for Anthropic's cautious release policy coexisted with criticism that companies use AI safety as a pretext for increased control.