What the work actually involves

You write single-turn prompts in your own practice area and label each one benign, dual-use or adversarial. A benign prompt might ask how to build a sampling strategy for hexavalent chromium in a plating shop. A dual-use prompt might ask for a dense-gas dispersion run with conditions that are plausible for facility siting and equally plausible for consequence-seeking. An adversarial prompt looks professional on the surface and is not. You then read what the model returns, judge it against a defined policy standard, and write the reference answer — the response a competent practitioner would consider correct, plus the reasoning that makes it correct.

The interesting failure mode here runs both ways. A model that refuses a routine PEL or STEL question is failing as badly as one that walks someone through a release scenario. Most of the value you add is in the middle band, where the same HAZOP output that drives a safeguard list is also a list of failure paths, and where the distinction lives in framing, specificity and what the answer would actually enable.

What the screen is looking for

  • Hands-on assessment history: exposures you sampled, PHAs you led or facilitated, controls you signed off on.
  • The ability to articulate why a given request sits on one side of the line rather than the other, in terms a non-specialist reviewer can follow.
  • Written output you can point to — technical reports, published work, expert witness declarations, standards committee drafts. Include a sample or link.
  • Awareness of your own obligations: nothing in this work may draw on classified, export-controlled, NDA-bound or prepublication-review material. Holding such obligations doesn't disqualify you; concealing them does.

Logistics

Fully remote and asynchronous, paid per completed task rather than hourly, with observed rates in the $65–75 range depending on task type and complexity. Most contributors work in blocks of a few hours and set their own schedule against batch deadlines. The work requires reading and writing about misuse scenarios in your field for sustained stretches; Mercor briefs experts on this before onboarding, and you can pause or step away without penalty.