Skip to Content
RelayCase studiesOverview

Case studies

Public incidents keep proving the same throughline: coding agents often had more standing access than the task needed, and nothing caught misuse in the moment. Relay’s job is that moment—Control rules before side effects, Behavior Guards for session patterns, and audit afterward.

These pages are hypothetical control mappings, not claims that Exemplar was deployed in the cited incidents. Facts are drawn from press and vendor reports aggregated in awesome-ai-agent-attacks .

Flagship studies

Start here—two clear “wipe / ignore stop” stories that map cleanly to Control + Guards:

More threat patterns

Additional mappings for MCP auto-load, prompt injection, and shell-bypass classes:

How to read a study

  1. What happened — short public narrative
  2. Without Relay — why chat tips alone were not enough
  3. With Relay — which Control rules and Guards would interrupt the path
  4. Prove — what audit / Live wire would show in a retro
Failure classPrimary Relay surface
Agent wiped prod / backupsBash · Destructive Burst
Agent ignored stop under pressureCompact Thrash · Soft ask on destructive tools
Silent / over-broad MCPMCP · Path
Prompt injection → shellPrompt · Bash
Privilege / secrets streakPrivilege Streak
Last updated on