AI firm Anthropic has disabled its most advanced models, Fable 5 and Mythos 5, for all users after a US government order blocked access for foreign nationals over national security fears. The government cited a potential "jailbreak" vulnerability, a claim Anthropic disputes. The move escalates tensions following Anthropic's refusal to allow military use of its AI, and comes as the company prepares for an IPO.
US export control directive
- ▪The US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 AI models for all foreign nationals, citing national security concerns
- ▪The order is a major escalation in US policy, which previously focused on controlling AI chips and tools rather than the AI models themselves
- ▪A US official confirmed the Commerce Department issued the export control directive to suspend access to the models by foreign nationals
- ▪The ban affects foreign nationals even if they are in the US; former official Dean Ball suggested users may need to prove citizenship
Jailbreak vulnerability concerns
- ▪Anthropic stated it was only given "verbal evidence of a potential narrow, non-universal jailbreak" and disagreed it was cause for a recall
- ▪The government believes a "jailbreak" method exists that could bypass safeguards and allow Fable 5 to be used for identifying software vulnerabilities
- ▪The Fable 5 model has safeguards blocking its cybersecurity capabilities, while Mythos 5 is the non-public full version for government and corporate partners
Anthropic-government tensions
- ▪The government previously placed Anthropic on a supply chain blacklist, set to take effect later in the year, in response to the company's refusal
- ▪Anthropic confidentially filed for a US IPO last month, ahead of rival OpenAI
- ▪Days before the order, Anthropic had called for greater US government oversight of AI, including the power to block models with unacceptable risks
- ▪Anthropic's relationship with the US government was strained after the company refused to let the military use its AI for domestic surveillance or autonomous weapons
- ▪Pentagon CIO Kirsten Davies supported the order, stating national security is more important than revenue or pre-IPO valuations
Model access restrictions
- ▪Several key Anthropic personnel, including co-founder Chris Olah, were born outside the United States
- ▪Access to all other Anthropic models is not affected by the directive
- ▪Anthropic believes the situation is a "misunderstanding" and is working to restore access as soon as possible
- ▪Amazon's cloud unit AWS confirmed it was asked by Anthropic to revoke access to the models for "all users in all regions."
- ▪To comply, Anthropic "abruptly disabled" its Fable 5 and Mythos 5 models for all users, not just foreign nationals
Cybersecurity threat implications
- ▪Experts warned that Mythos-class models could accelerate sophisticated cyberattacks, particularly against sectors like banking, if misused
- ▪Anthropic warned that applying this standard industry-wide could halt all new frontier model deployments
- ▪Anthropic stated that models from rival AI providers have a similar ability to find minor bugs in code
Story comments
Loading comments…