C.W.K.
Stream
Lesson 09 of 10 · published

Incident Response for AI Systems

~14 min · production, incidents

Level 0Apprentice
0 XP0/100 lessons0/14 achievements
0/120 XP to next level120 XP to go0% complete

The 2 a.m. page

Something is wrong: outputs got worse, cost spiked, an injection succeeded, a tool fired the wrong action. Incident response for AI systems borrows from classical SRE, with three additions: prompt provenance, model attribution, and trace forensics.

The first ten minutes

  1. Identify the affected prompt version and model from the metric tag.
  2. Roll back the version flag (the flag exists, right?).
  3. Confirm rollback served traffic by metric tag count.
  4. Snapshot logs and traces for forensics.
  5. Communicate: status page or internal channel.

The next hour

  • Pull the bad traces; identify failure mode.
  • If it's a provider issue, check status pages and consider fallback chain activation.
  • If it's a prompt regression, identify what changed in v_new vs v_old.
  • If it's an injection / abuse, identify the entry point and add a defense layer.

Post-mortem

Write it. Convert findings into eval cases. Add new dashboards or alerts that would have caught this earlier. Don't blame the model; blame the absence of guardrail / eval / canary.

Code

Pre-built rollback runbook (sketch)·markdown
# AI prompt rollback runbook

1. Open feature flag dashboard.
2. Set `prompt_v9` flag to 0%.
3. Verify in metrics: `prompt:v8` count > `prompt:v9` count within 5 minutes.
4. Snapshot traces for `prompt:v9` from last hour.
5. Create incident channel; invite owner + responder.
6. Within 2h: post initial findings.
7. Within 24h: post-mortem doc with eval-set additions.

External links

Exercise

Run a rollback drill on a non-critical prompt. Time the steps. Identify the slowest step and improve it.

Progress

Progress is local-only — sign in to sync across devices.
Spotted a bug or have feedback on this page?Report an Issue

Comments 0

🔔 Reply notifications (sign in)
Sign inPlease sign in to comment.

No comments yet — be the first.