level: business
the us government on friday ordered anthropic to immediately shut off access to claude fable 5 and claude mythos 5 for all users worldwide. the directive, framed as an export control action, came after the government claimed a potential jailbreak of fable 5. anthropic complied but disagrees with the decision, arguing that the jailbreak is narrow and not universal, and that similar capabilities exist in other public models like openai's gpt-5.5.
mythos is anthropic's most capable model, kept restricted due to its ability to find security flaws in software. it was shared with about 50 vetted organizations for defensive cybersecurity. fable 5, released three days earlier, was a version with guardrails blocking high-risk responses, making it publicly available. benchmarks showed it was the most capable public model at the time.
anthropic says its safeguards use independent classifier systems that remain effective even if a model is prompted past a refusal. the company found no evidence of those safeguards being bypassed to produce harmful content. anthropic warned that applying this standard across the industry would halt all new model deployments. the situation highlights the tension between safety messaging and regulatory action, as anthropic's own caution about mythos may have drawn government scrutiny.
why it matters: this shows how safety claims can trigger regulatory intervention, affecting ai deployment and access for data science and cybersecurity work.