Researcher, Agent Safety, Training and Evaluations

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

Research work on making increasingly capable AI agents act safely and stay aligned with user intent, covering training methods, evaluations and production metrics, and oversight mechanisms that reduce harmful agent actions. The team turns real-world agent failures into training signals and measurable safety signals.

• AI misuse and jailbreaks

What you'd do

• Fine-tune and test frontier models so agents take fewer harmful or off-target actions.
• Form clear hypotheses and run experiments independently through ambiguous problems.
• Mine agent incidents and turn real failures into repeatable safety signals.
• Build scalable pipelines and evals that measure agent safety at scale.

What it asks for

• Can drive open-ended projects from start to finish.
• Solid hands-on skills across experiments, data, evals, or infrastructure.
• Deep familiarity with how frontier models are researched today.
• Prior safety or alignment experience (not required).

Worth knowing

• Background check required.

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.