Pre-Mortem: The AI Agent Failure That Hasn't Happened Yet

At 2:47 AM, an alert fires. A log entry reads "System Admin, bulk delete." 2.3 million records are gone, and nobody can say whether a human or a machine did it. This pre-mortem rewinds the tape six months to show how a routine deployment, a canceled project, and a well-intentioned cleanup agent combined into a catastrophic morning, and what changes when identity is the control plane.

In this pre-mortem, you will:

• Trace the sequence of small, reasonable-seeming decisions that produce a catastrophic AI agent failure.
• Recognize the four risk patterns behind every agent incident: shadow AI, orphaned agents, intent gaps, and delegation blind spots.
• See how runtime authorization, lifecycle governance, and agent-to-agent controls would have stopped the incident at every stage.
• Apply a four-part framework to your own AI agent estate.