Cathedral, founded by former members of Musk's DOGE, used an AI Agent to breach Anthropic's internal network in 12 hours during a bug bounty test

According to The Washington Post, Cathedral, an AI cybersecurity startup founded by former members of Musk's DOGE project, deployed its own AI Agent during Anthropic's bug bounty test. After several days of frontal assault on its own software failed, it pivoted to vulnerabilities in third-party commercial software used in Anthropic's daily office work, and within less than 12 hours opened a path into internal sensitive data and communication systems. Cathedral's draft report said the Agent saw meeting minutes, employee salaries and hiring offers, and also entered a Slack channel where employees discussed unreleased frontier models. Anthropic acknowledged that researchers accessed limited sales, HR and IT systems, but stressed that user Claude chat logs, source code and model weights were not touched, and accused Cathedral of exaggerating the impact and not fully complying with bounty rules, saying its monitoring system had detected anomalies before the report was submitted, cut off access that day and disabled the software involved. Cathedral denied violations, saying data access was limited to the minimum necessary for security research. The company, founded only months ago, has raised $160 million at a $1.4 billion valuation.