White House pushes back on Anthropic’s plan to broaden Mythos AI model access: Report

The White House is pushing back on a proposal from Anthropic to widen access to its artificial intelligence (AI) model Mythos, a tool capable of launching complex cyberattacks, The Wall Street Journal reported on Wednesday.

Around 50 companies and organizations tied to critical infrastructure currently have access to the model, but Anthropic has suggested expanding access to about 70 more, the report said.

Administration officials opposed the idea over security concerns and worries that Anthropic may lack the computing capacity to support additional users without limiting the government’s own access.

The White House is closely involved in the rollout of Mythos due to national security risks posed by the model, the report said, adding that its ability to detect and exploit software vulnerabilities has alarmed government agencies and businesses.

Anthropic said Mythos has found "thousands of high-severity vulnerabilities, including some in every major operating system and web browser," which could be exploited by malicious actors.

“The fallout – for economies, public safety, and national security – could be severe,” it said, adding that the initial group of organizations with access is using the model to strengthen defenses and share insights.

Ongoing talks between Anthropic and US officials aim to manage those risks and repair their strained relationship, according to The Wall Street Journal.

The administration previously designated Anthropic a "supply chain risk" after it refused Pentagon demands for unrestricted access to its AI models, specifically barring use in fully autonomous weapons or mass domestic surveillance. Anthropic has filed two related lawsuits.