AI security startup Cathedral breached Anthropic's internal systems in under 12 hours using AI agents

According to a Washington Post exclusive, Cathedral, an AI cybersecurity startup founded by former DOGE members, revealed that during an authorized security test against Anthropic, its researchers used AI agents to find a path into Anthropic's sensitive internal systems in under 12 hours, obtaining a trove of sensitive documents — including Slack channels discussing the most sensitive model information, meeting notes, and employee data such as salaries and offer letters. Polymarket also posted the breaking news.

Confirmed

• Cathedral informed Anthropic of the successful breach; the intrusion took place within the framework of an Anthropic-authorized security testing program
• The penetration took under 12 hours and involved autonomous execution by AI agents
• Accessible data included sensitive Slack channels, meeting notes, salaries, and offer letters

Unconfirmed

• Technical details of the intrusion path: @AlexTensor reshared claims about how the breach was achieved, but the materials lack complete, reliable technical detail
• Follow-up handling and remediation between Cathedral and Anthropic

Why it matters

• The incident came shortly after an Anthropic whistleblower publicly criticized the company for keeping highly sensitive work on Slack, echoing that criticism
• Anthropic, a company built around AI safety, was rapidly breached by AI agents — highlighting the asymmetric risks of AI-driven offense versus defense, and demonstrating the real-world power of AI agents in red-team security testing

2026-10-26 ~ 2026-10-26 · 7 related posts

• AI Agents Breached Anthropic Internal Systems in Under 12 Hours, WaPo Reports — connoraxiotes · 2026-10-26
• Ex-DOGE cyber researchers breach Anthropic systems with AI agents in under 12 hours — ScottNover · 2026-10-26
• [source] Ex-DOGE staffers' startup Cathedral says it broke into Anthropic in 12 hours — AlexTensor · 2026-10-26
• [source] Ex-DOGE staffers' AI security startup breaches Anthropic's internal systems in under 12 hours using AI agents — Polymarket · 2026-10-26
• Polymarket pegs Anthropic training pause this year at just 7% amid breach claim — Polymarket · 2026-10-26

2 near-duplicate retellings: eliano · stanfordnlp