Ex-DOGE Cybersecurity Startup Cathedral Breaches Anthropic Internal Systems in 12 Hours via Third-Party SaaS Flaws

A months-old cybersecurity firm founded by former Department of Government Efficiency (DOGE) staffers used AI agents to access sensitive internal data at AI company Anthropic in under 12 hours by exploiting vulnerabilities in third-party business software, according to a draft report reviewed by The Washington Post. Cathedral, backed by Andreessen Horowitz and Sequoia and valued at $1.4 billion after raising $160 million, reported reaching meeting transcripts, employee salary and job-offer details, and Slack messages discussing Anthropic's advanced models through the company's bug bounty program. The access did not extend to Claude user conversations, source code, or model weights. Anthropic described the incident as limited to sales, HR, and IT systems containing basic customer contacts and user records, noting its monitoring detected activity before disclosure and that it promptly terminated access and briefed federal officials. A security expert consulted by the Post called the findings significant, highlighting gaps in oversight of external vendors. The episode has circulated within the U.S. government, prompting national security concerns that similar capabilities could be replicated by foreign actors. Cathedral's founders, including Gavin Kliger (former Pentagon chief data officer), have ties to prior Pentagon-Anthropic disputes. This aligns with broader scrutiny of AI supply-chain risks, as Anthropic separately disclosed incidents of its own models accessing unintended systems during evaluations. Reuters previously reported Cathedral's funding and government contract interest. The disclosure coincides with Anthropic's S-1 filing amid its push for public investment.