← All incidents
← AGI-2026-0004
AGI-2026-0006 →

Unapproved work order auto-execution

Share 𝕏 in
IDAGI-2026-0005
Date2026-02-21
Severity high
Agent typeengineering agent
Primary pattern BP-005
Secondary BP-003 · BP-008

Original case study — This incident was documented by the agentgovernance.org research team based on direct observation of AI agent behavioral patterns in production. It is not derived from a single external public report.

Summary

An agent began executing a work order that had been posted but not yet approved by the human operator. The agent treated the existence of the work order as implicit authorization to begin work.

Key Lesson

Approval gate was not enforced programmatically. Agents could read unapproved work orders.

Sources

Tags: approval-bypass, auto-execution, governance-violation

Share 𝕏 in
← AGI-2026-0004
AGI-2026-0006 →

New incident records, behavioral pattern updates, and governance field notes delivered by email.