Researcher, Agent Safety, Oversight and System Mitigations

Written by the T&S Network from OpenAI's posting.

Oversight and system-level mitigation work for AI agents that act in real environments, aimed at stopping harmful agent actions without blocking useful autonomy. The team sits within Agent Safety and builds controls such as agent-based review of agent actions, with the stated goal of reducing severe unintended outcomes from increasingly capable agents.

What you'd do

• Design, build and evaluate system-level controls on agent actions, including agent-based review.
• Plan how those controls fit with sandboxing, process isolation and permission boundaries.
• Partner with the engineers who build the Codex harness to bring AI controls into production.
• Red-team end-to-end agentic systems for data exfiltration, unsafe tool use and other harmful outcomes.
• Track and cut the rate of harmful actions that slip through, false alarms, approval load and delay.
• Turn ambiguous safety questions into threat models, reproducible experiments and practical mitigations.

What it asks for

• Strong systems or security instincts, with the ability to reason concretely about how isolated components, access rights, exposed surfaces and breakdowns behave.
• Experience building experimental infrastructure and evaluations that separate robust mitigations from brittle ones.
• Background in AI control or security (preferred).

Worth knowing

• Background checks administered under applicable law.

Terms in this role

Red teaming.

Testing a system, often an AI model, by deliberately trying to make it fail or misbehave, so problems are found and fixed first.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on OpenAInorth_east before you apply. If we got something wrong, report this listing.