Read, Watch or Listen

WASHINGTON — The U.S. Cybersecurity and Infrastructure Security Agency (CISA) is using Anthropic’s Mythos artificial intelligence model to audit government software code, according to three people familiar with the project. CISA’s Attack Surface Evaluation team is scanning government code repositories with Mythos to detect bugs and security flaws that could be exploited by foreign intelligence services or cybercriminals. Two sources said the reviews have already uncovered a large number of vulnerabilities, though they did not describe the specific systems affected or the nature of the weaknesses identified in the government software. WASHINGTON — The deployment of Mythos by CISA comes despite an ongoing standoff between Anthropic and the White House over AI safety safeguards. In February, Anthropic refused to remove protections that prevent its systems from being used for autonomous weapons or domestic surveillance, prompting the Pentagon to issue a formal supply‑chain risk designation, a label previously reserved for foreign companies suspected of espionage, before a judge blocked the effective blacklist in March. The dispute eased after Anthropic privately released Mythos, which U.S. officials, including analysts at the National Security Agency who have used it since April in classified settings, view as highly capable at identifying and exploiting cybersecurity vulnerabilities. The clash intensified again when the White House pushed for restrictions on foreign access to a public, safety‑enhanced version of the model called Fable, leading to a temporary global shutdown that ended last week, as Anthropic confidentially prepared for a U.S. initial public offering.

Prepared by Jonathan Pierce and reviewed by editorial team.