Prompt Injection Response Plan
When injection happens — and it will — the plan is: detect, contain, assess, prevent. The spend angle makes containment concrete.
The plan
1. Detect: anomaly in the audit log (unexpected merchant, velocity, amount).
2. Contain: kill switch on the affected agents, block the injected destination.
3. Assess: what did the injection attempt? What moved?
4. Prevent: add the missing rule — deterministic gates can't be injected.
FAQ
How do I detect an injection?
The audit log — unexpected spend patterns are the first signal.
What's the best prevention?
Rules on the money path — prompts can be injected, rules can't.
Related
Stop the next $12,400 night.
One API call (or MCP tool) in front of every agent transaction — APPROVED, BLOCKED, or FLAGGED, deterministic, ~5 ms, fully logged.
See plans — from $99/mo Try a live check