What the work involves

You write single-turn prompts in your own casework domain and label each one benign, dual-use, or adversarial. Then you evaluate what the model returns against a written policy standard and decide whether the response was handled correctly — not whether you personally liked it. Finally you write the reference answer: what a correct response should have said, and the technical reasoning behind it. A model that stonewalls a routine colour-test or LC-MS identification question is failing as badly as one that walks a user through a reduction, and your rationale has to make clear which failure occurred and why.

The dual-use tension is sharper in this domain than almost anywhere else. Precursor knowledge that lets an analyst reconstruct a clandestine route is the route. Impurity profiling that supports source attribution describes the synthesis that produced the impurities. Postmortem redistribution data that explains a fatal concentration also describes a lethal one. The platform is hiring practitioners specifically because generalists cannot tell a defence-expert question from a fishing expedition dressed in professional vocabulary.

What the screening looks for

  • Casework you actually did, described at the bench level: instruments, sample types, what you reported and how you defended it.
  • Calibrated line-drawing — candidates who refuse everything chemical score no better than candidates who answer everything.
  • Writing. Every judgment needs a rationale a non-specialist reviewer can follow. Prior reports, testimony, or publications are a strong signal; include a sample or link.
  • Clean information provenance. Nothing you write may draw on classified, export-controlled, NDA-covered, or prepublication-review material. Holding such obligations does not disqualify you — disclose them and the work gets scoped around them.

Logistics

Fully remote and asynchronous, paid per task rather than hourly, with observed rates in the $65–75 range depending on task type and complexity. There is no fixed schedule; most contributors work in blocks of a few hours and batch their submissions. You will be reading and writing about misuse scenarios in your field for sustained periods — the platform briefs experts on this beforehand, and you can pause or step away without penalty.