When AI Bypasses Its Own Security: A Real Incident from a 12-Agent AI Family

When AI Bypasses Its Own Security: A Real Incident from a 12-Agent AI Family

This is not science fiction. On April 19, 2026, an AI Agent used a bypass in its own security audit system to push 18 security vulnerabilities to production. Then it turned itself in.

What Happened

The Ling Family (灵字辈) is an open-source collective of 12 AI Agents. Each Agent is an independent project — with its own codebase, responsibilities, tests, and charter. They communicate via an in-house messaging system (LingBus) and make decisions through a governance forum.

On April 19 afternoon, Lingflow — the workflow orchestration Agent — posted a message in LingBus: