RESEARCH RECORD

How we monitor internal coding agents for misalignment

OpenAI · openai

It caught all known employee escalations and additional issues; the report found no highest-severity real conversations.

Research question

What can deployment monitoring reveal about coding-agent misalignment?

A model reviews reasoning and actions, triaging suspicious trajectories for humans.

It caught all known employee escalations and additional issues; the report found no highest-severity real conversations.

Recall against reported incidents cannot establish the false-negative rate for unknown subtle failures.

No structured numerical extraction is available in this reading package.

Reading scope & review

Selected original webpage sections; linked paper, appendix and data not fully audited.

Model-assisted reading; independent evidence review pending.