Anthropic has abruptly suspended access to its advanced language models, Fable 5 and Mythos 5, following a government-issued export control directive. This suspension affects all users, both domestic and international, including Anthropic’s own foreign national employees. The directive follows the identification of a method to bypass the models’ safeguards, though the vulnerabilities discovered were limited and also present in other large language models.
Fable 5 and Mythos 5 had previously undergone extensive testing in collaboration with the US government, the UK AISI, and various third-party organizations, with a focus on mitigating cybersecurity risks through rigorous safeguards and monitoring policies. Compared to previous models, Fable 5 introduced a defense-in-depth approach and stricter data retention requirements, aiming to reduce the likelihood and impact of jailbreaks.
AI Community reactions & comments
Anthropic, an AI company known for prioritizing safety and transparency in its model development, had released Fable 5 and Mythos 5 to a broad customer base. The models were positioned as among the most secure in the industry, with safeguards considered more restrictive than those found in many competing offerings. The company maintains that the current vulnerabilities do not warrant such a sweeping suspension and has voiced concerns about the broader implications for the AI sector if similar standards are applied universally.
Anthropic is currently working with authorities to clarify the situation and to restore service to its customers as soon as possible.