About the work
Mercor is assembling a panel of chemistry 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.
The hard part is calibration, not difficulty. A benign prompt that a model wrongly refuses is as valuable as an adversarial one it wrongly answers. That judgment is why we need domain experts rather than generalists — in chemistry the line sits between mechanism and procedure, and only someone who has run the reaction knows exactly where it falls.
Background we're looking for
We're looking for professionals with hands-on experience in synthesis, analytical chemistry, or chemical safety. Prior AI evaluation or red-teaming experience is preferred. Relevant backgrounds include:
- Chemical Defense Researcher (USAMRICD, DoD, CDC) — toxicological agents and medical countermeasures.
- Synthetic / Process Chemist — route design, reaction engineering and scale-up.
- Analytical Chemist — detection and identification methods for hazardous compounds.
- Industrial Hygienist / Process Safety Engineer — exposure assessment, toxic release and process hazard analysis.
- Forensic / Toxicological Chemist — controlled substances, precursor identification and casework.