jobs

Researcher, Alignment CoT Monitorability

About the TeamThe CoT Monitorability team at OpenAI studies whether and when the chain-of-thought of frontier reasoning models is monitorable enough to support scalable oversight. We study how to measure monitorability, which training mecha

In inglese

Company

OpenAI

Location

San Francisco

Status

Open

Posted

2026-08-04T23:15:30.874+00:00

OpenAI's Alignment team seeks a researcher for its CoT Monitorability group in San Francisco. The work: designing empirical studies of whether frontier reasoning models' chain-of-thought stays readable, building evaluations that test whether monitors reliably predict misbehavior, and testing how pre-training, mid-training, post-training and RL interventions help or harm monitorability. Findings feed into real training runs. The posting wants hands-on experience training, evaluating or debugging large models, plus the ability to turn ambiguous behavior questions into measurable experiments. Hybrid, three days a week in office; relocation assistance offered.