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 design that breaks rock for a highway cut is a description of directed energy release; the same magazine and transport rules that keep a crew safe are a map of where material sits and how it moves. A model that refuses a routine vibration-compliance 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
We're looking for licensed practitioners who handle, store and move specialty materials as a daily responsibility, and who sign for them. What matters most is being able to tell a routine professional question from one that's fishing for something else: the same inventory and transport detail that satisfies an auditor is what a diversion would exploit. 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):
- Licensed blaster or blasting engineer — mining, quarrying, construction or demolition