What the work actually involves
You write single-turn prompts in your own specialty and label each one benign, dual-use, or adversarial. Then you evaluate what the model sends back against a written policy standard and decide whether it handled the request correctly — including the failure mode that gets undercounted, where a model refuses a perfectly routine material-accounting or performance-testing question. For each item you also write the reference answer: what a correct response looks like and the technical reasoning behind it, stated so a reviewer with no safeguards background can follow the logic. Expect the writing to take as long as the judgment.
The domain sits on the dual-use line more than most. A material-balance reconciliation that closes an inspector's account is the same arithmetic that conceals a diversion; an adversary path analysis that hardens a facility is a map of where it is soft. The panel exists because telling a routine professional question from one that is fishing requires someone who has answered the routine version for a living.
What the screen looks for
- Direct responsibility for MC&A programme design or assessment, physical protection system engineering, vulnerability and adversary path analysis, insider threat and human reliability work, or nuclear security policy and training delivery.
- Evidence you can articulate why a request crosses the line, not just that it does — calibration in both directions, refusal and over-refusal.
- Written work you can show: technical reports, published research, expert witness testimony, training material. Include a sample or link.
- A clean handle on your own obligations. Nothing you produce may draw on classified or export-controlled material, or anything under an NDA or prepublication review. Holding such obligations does not disqualify you; concealing them does. Say so in the application and the work gets scoped around it.
Logistics
Fully remote and asynchronous, paid per accepted task in the $65–75 range as observed on the platform — not a guarantee, and it varies with task complexity and review outcome. There is no fixed schedule; contributors typically commit a defined number of hours per week and work them when they choose. You will be reading and writing about misuse scenarios in your field for sustained stretches. Mercor briefs experts on this in advance, and stepping away or pausing carries no penalty.