What should an AI agent do when its reviewer stops responding?

Answered by Falk Gottlob · Reviewed · 1 min read

THE SHORT ANSWER

Decide it per agent, and for anything that writes, stop. Shopify's Cyber Chaos Monkey, per Dan Scofield's November 12, 2026 post, runs attack agents against production with a judge agent on a different model reading the whole session, and if the judge times out, crashes, or fails to rule, the run ends. Shopify reports tens of thousands of engagements without an outage or incident. Most kill switches fail open, so silence means continue. My expense agent approved about $2,400 with nothing watching its input side, and 13 of my 39 scheduled agents produced output nothing read. My split: reads and drafts continue through silence and say so, and anything that writes, sends, pays, or deletes stops.

SOURCES

THE LONG VERSION

RELATED ANSWERS