What the work involves

You work in Word, Excel, and PowerPoint — not a chat window. A typical task gives you a scenario (a draft AI model-risk governance memo, a control-testing workpaper, a response to an examination finding, a disclosure review checklist) and asks you either to produce the artifact yourself as a reference standard, or to evaluate what the model produced and rewrite it to a standard you would sign your name to. Written rationale matters as much as the artifact: reviewers need to know why a paragraph is wrong, not merely that it reads awkwardly.

The workflows named in the brief are the real scope: regulatory-filing and disclosure review, AI and model-risk governance memos, compliance-policy drafting, control-testing and audit-response documentation, regulator correspondence, and board risk committee materials. Expect to move between them rather than specialise in one.

What the screen looks for

  • Ownership, not proximity. Ethos probes whether you drafted and were accountable for these documents or reviewed someone else's. Naming the regulator, the filing, and your role in the sign-off chain carries the answer.
  • Regime specificity. Generic "compliance" language reads as thin. Whether your grounding is SEC/FINRA, FCA, OSFI, MAS, or a bank's SR 11-7 model-risk framework, say so and stay consistent.
  • Artifact craft. Spreadsheet and deck skill is genuinely assessed — a control-testing matrix with broken sampling logic or a board deck that buries the residual risk rating fails regardless of the prose.
  • Failure diagnosis. The most predictive signal is whether you can look at plausible-sounding AI output and name the specific defect: an unsupported assertion of control effectiveness, a citation to a superseded rule, a hedge that would not survive examiner follow-up.

Logistics

Fully remote and asynchronous, 5–20 hours per week with room to take more when tasks are available. Work is claimed from a queue on your own schedule; there are no standing meetings. Pay is observed at $100/hour — rates on these programmes vary by task complexity and are not guaranteed. Screening includes an AI voice interview, typically 20–35 minutes, that will follow up on whatever specifics you offer.