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 propulsion analysis that qualifies a motor for flight describes a capability; the same safe-and-arm logic that keeps a system inert is the part someone would want to defeat. A model that refuses a routine performance or qualification 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 engineering half of the domain: we're looking for people who have designed, qualified or tested propulsion and initiation systems, with real hardware and real test data behind them. What matters most is being able to tell a routine professional question from one that's fishing for something else: the same effects model that sizes a safety standoff also sizes an intent. 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):
- Munitions or warhead design and effects engineering