Why it matters
METR outlines how an assessor could examine monitoring of internal AI agents. The method reconciles inference and monitoring logs, checks whether agents can alter controls, samples unflagged activity, and tests escalation to people. It includes preprocessing and context compaction in the assessed pipeline. This is a proposed assessment method, not evidence that any particular monitor reliably detects dangerous behavior.
My takeaway: In an authorized test environment, reconcile parent and child model calls against monitor records. Introduce harmless cases representing a missing child trace and a disabled monitor, then check detection and escalation time. Sample unflagged runs, inspect truncated inputs, and follow a test alert through the actual human response process.