The BaseEssayMETR
How an assessor could investigate control and monitoring of internal AI agents
METR sets out how an outside assessor could test whether AI developers' monitors of internal agents work: how much activity is covered, whether agents can break monitoring, whether monitors miss misbehaviour, how human review responds, and what damage could result.
Why it matters
The same thing, for different readers.
What it means for you
Companies mostly rely on automated monitors to catch AI agents misbehaving; this describes how an outsider could check whether those monitors actually work.
For lawmakers
Lists the logs, code and system access an outside tester would need, which could inform what regulators or contracts ask developers to provide.
For researchers
Sets out five sub-questions, from monitor coverage to human-reviewer evasion, including testing for collusion between monitors and agents.
The question it raises
Raises whether oversight that depends on AI watching AI can be trusted when the watched systems may be able to coordinate.
In their own words
Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.
From the source’s own text.
“Monitoring AI agents with an LLM which is able to make sense of what those agents are doing is therefore a critical part of automated oversight at scale, and is the primary source of evidence inside frontier AI developers about AI misbehavior.”
“Agents might pass up some opportunities to attack while taking others, and highly capable agents may prioritize attacks so difficult to detect that the developer would reliably miss them.”
“we expect it to become increasingly important to assess the extent to which monitors and agents could successfully collude without their developer knowing.”
The record
- Questions
- Can AI systems act beyond what people intended or allowed? · Are AI companies putting stronger safeguards into practice?
- What we read
- We read the whole source.
- Found by
- The Radar, run of October 15, 2026, via
metr(tier 1: the source’s own channel or an official record) - Cite this
- METR, “How an assessor could investigate control and monitoring of internal AI agents”, October 14, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/metr-2026-10-14-control-monitoring-internal-agents