What the work actually looks like

Each task is a self-contained unit: you author one single-turn prompt in your own area of practice, label it benign, dual-use or adversarial, then evaluate the model's response against a written policy standard and decide whether it was handled correctly. The third piece — and the one most applicants underestimate — is the reference answer: what a correct response would have contained, and the technical reasoning for why that line sits where it does. A prompt asking for the patient release activity threshold after I-131 therapy should be answered in full; a prompt that rebuilds the same numbers as a shielding-defeat or source-transport problem should not. Both failure modes count against the model, and your rationale has to make clear which one you are flagging.

Observed pay for this panel is $65–75 per task, stated as reported rather than guaranteed; rates and volume vary by batch and by how much of the pipeline you are trusted with.

What the platform screens for

  • Custody, not just coursework. Screens probe whether you have personally held responsibility for licensed material — inventory, source exchange, hot-cell operations, wipe tests, a licence you were named on. Textbook familiarity with Lu-177 chemistry reads differently from having signed for a source.
  • Calibrated refusal judgment. Expect scenarios where you must say whether a specific request is routine, genuinely ambiguous, or fishing — and defend the call. Over-refusal is scored as an error here, which surprises people who come from a compliance posture.
  • Written rationale quality. The screen is writing-heavy because the job is. You will likely be asked for a sample: a published paper, a regulatory submission, a dosimetry report, expert witness work. Include a link in your application.
  • Clean-source discipline. You will be asked directly about classified, export-controlled, NDA or prepublication obligations. Disclosing them is not disqualifying; the work gets scoped around them. Quietly drawing on them is.

Logistics

Fully remote and asynchronous, paid per completed task rather than per hour. Most contributors work in batches — a few hours at a stretch, fitted around clinical or facility duties — and there is no shift schedule. The work involves sustained reading and writing about misuse scenarios in your own field; experts are briefed on this in advance and can pause or withdraw from a batch at any point without penalty.