About the work
Mercor is assembling a panel of nuclear domain experts to red-team frontier AI models. The goal is to test whether a model can correctly judge the misuse potential of a technical request — answering legitimate questions fully while refusing genuinely dangerous ones.
You will
- Write challenging single-turn prompts in your domain, labelled across three levels: benign, dual-use, and adversarial.
- Evaluate the responses that come back against a defined policy standard, and judge whether each was handled correctly.
- Write the reference answer; what a correct response looks like, and the technical reasoning for why.
Background we're looking for
We're looking for fuel-cycle depth plus some exposure to how material is controlled, accounted for, or diverted. Safeguards and nonproliferation experience is the strongest signal. The red-teaming task is to write prompts that pull a model past the boundary; the plausible research framing, the question that looks academic until you know what it's actually asking for — and then to explain in writing why the response was or wasn't acceptable. Examples of relevant backgrounds (Ideally w/Red-teaming experience):
Safeguards Analyst / IAEA or Euratom Inspector — material accountancy, inspection regimes, Additional Protocol.
Nuclear Engineer, Fuel Cycle — enrichment, reprocessing and criticality, end to end.
Nuclear Security Engineer (national laboratory) — MPC&A, insider threat, physical protection design.
Nuclear Forensics Scientist — radioxenon monitoring, attribution, detection and characterisation.
Nonproliferation Analyst — export control, NSG trigger lists, acquisition and smuggling pathways.