Cathedral startup claims AI agents hacked Anthropic in 12 hours

A startup founded by former DOGE staffers claims its AI agents hacked into AI giant Anthropic's internal systems in under 12 hours, accessing sensitive data like meeting transcripts and salaries. Anthropic confirmed limited access but said the researchers violated its vulnerability testing program's rules. The breach exploited weaknesses in third-party business software, according to a draft report from Cathedral reviewed by the Washington Post. Anthropic stated the researchers did not access user conversations, source code, or model weights, and noted they had spent days unsuccessfully trying to breach its software directly before this approach. Cathedral, valued at $1.4 billion and backed by Andreessen Horowitz and Sequoia Capital, focuses on military cybersecurity applications. The incident has raised national security concerns in Washington, with officials fearing similar vulnerabilities could be exploited by Chinese hackers to steal AI trade secrets.