OpenAI hired the hackers who broke into Anthropic

OpenAI said Monday it will bring in Cathedral Systems, the security startup whose AI agents found routes into parts of Anthropic's internal systems last month, to attack its own defenses in its next internal security sprint, using OpenAI's models alongside its own teams.

Cathedral, founded by four former DOGE staffers, says its agents reached Anthropic Slack messages, meeting transcripts and salary records in under 12 hours through flaws in business software, not Claude itself; Anthropic says the access was limited to sales, HR and IT systems and accuses Cathedral of breaking its bug-bounty rules.

OpenAI is also paying up for holes: typical rewards for serious bugs more than quadruple, with exceptional critical finds worth up to $172,000, after Reuters reported employees had warned about OpenAI's own security lapses since before ChatGPT launched.