Anthropic disables advanced AI models after government order
Quick Look
- US AI company Anthropic has disabled access to its advanced AI models, Fable 5 and Mythos 5, following a government order to suspend access by foreign nationals.
- The company is assessing compliance with the export-control order, which cited potential security risks related to bypassing safety controls.
AI-generated summary
Why It Matters
Anthropic, a US AI company, disabled its most advanced AI models, Fable 5 and Mythos 5, after receiving a government order to suspend access by foreign nationals. The order cited potential security risks related to users bypassing the models' safety controls.
US AI company Anthropic said on Friday that it disabled access to its most advanced AI models, Fable 5 and Mythos 5, after receiving a government order to suspend access by foreign nationals.
The company said it shut down the models for all users while it assesses how to comply with the export-control order. The order also applied to foreign-national Anthropic employees, it added.
The company said the government did not provide detailed information about the alleged security risk. According to Anthropic, officials believe users may have discovered a way to bypass the models’ safety controls and get them to help find software vulnerabilities.
Anthropic disputed the significance of the bypass, saying it was only able to uncover a small number of previously known, minor software vulnerabilities.
Anthropic released Fable 5 and Mythos 5 earlier this week. Fable 5 was intended for general use with additional safeguards, while Mythos 5 was only made available to a limited group of trusted cybersecurity partners with some restrictions removed. Anthropic said Fable 5’s capabilities “exceed those of every model we’ve previously made generally available.”
On Wednesday, Anthropic said it would make restrictions governing its most advanced AI models more transparent, including by disclosing when user requests are downgraded or rejected. The move came after criticism that some limitations were not visible to users.
Before the policy change, Anthropic could silently route requests involving cybersecurity, biology, and advanced AI development from Fable 5 to the less capable Opus 4.8 model. Under the new policy, users would be notified when a request is flagged, while Application Programming Interfaces (API) developers would receive explanations for any rejection or fallback to another model.
The approach drew criticism from researchers, who argued that routing some requests related to frontier AI development to a less capable model could slow progress in the field. Responding to the backlash, Anthropic agreed to make the safeguards visible.
The company previously cited national security concerns as a reason for rejecting or downgrading some requests, arguing that foreign adversaries could use its technology to strengthen their AI capabilities.
Open Questions
- What specific security risks are alleged?
- What are the long-term implications for AI development and export controls?
