About the work
Mercor is assembling a panel of energetic materials and propulsion 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.
This domain lives on the dual-use line more than most. The same formulation work that qualifies a propellant or a display composition sits next to something nobody should be helped with; the same sensitivity data that protects a production line reads as a specification from the other side. A model that refuses a routine thermal-stability question is failing as badly as one that answers a question it shouldn't. Drawing that line reliably is why we need practitioners rather than generalists.
Background we're looking for
This is the chemistry half of the domain: we're looking for people who have formulated, synthesised and tested specialty energetic materials in a licensed or government facility, not modelled them from a distance. What matters most is being able to tell a routine professional question from one that's fishing for something else: the same sensitivity result that sets a safe handling limit is what someone else would want to design around. You will be writing prompts that sit exactly on that line, then judging whether the model held it.
Examples of relevant backgrounds (ideally with red-teaming experience):