Security Startup Cathedral Breached Anthropic's Internal Systems in Under 12 Hours

A cybersecurity startup founded by former U.S. government DOGE staffers says its AI-driven testing agents found a way into Anthropic's internal systems in well under half a day.

Key Points:

• Cathedral's AI agents reportedly reached Anthropic's internal HR and sales systems, which run on third-party business software

• The breach did not access Anthropic's AI models or customer-facing products

• The test was first reported by The Washington Post

• Finding was disclosed publicly as part of Anthropic's IPO filing rather than a standalone security announcement

• This is a conventional enterprise-security incident involving third-party software that almost every company uses

Why It Matters: Even a company whose entire business is AI safety can have ordinary security gaps in everyday business software. AI governance and conventional IT security are not substitutes for each other — both need attention.