What the work actually involves
AI labs need compliance work product they can grade against. In practice that means you'll be asked to do things like draft a realistic scenario — a sanctions screening hit, a SOX control deficiency, a third-party risk questionnaire with gaps — and then produce the answer a competent practitioner would give, with the reasoning made explicit. Other tasks run the other direction: a model produces a risk memo or a control narrative, and you score it against a rubric, flag where it invented a regulatory citation, and write the correction. Expect a mix of task authoring, side-by-side response comparison, and written critique. The deliverable is usually prose plus structured judgments, not code.
What the screen looks for
Mercor's AI interview is conversational and follows up. It probes whether your resume claims hold under pressure — which frameworks you actually operated under, whether you owned a control or observed one, what your escalation call was when a finding was ambiguous. Vague answers about "managing compliance programs" get drilled. The other thing it measures is evaluation instinct: can you say why one answer is better than another in terms someone else could apply, rather than just asserting it. Regulatory precision matters — confidently misstating a rule or a threshold is a fast disqualifier, since catching exactly that failure mode in models is the job.
Logistics
- Fully remote, largely asynchronous, with task queues rather than fixed shifts.
- Typical engagements run 15–30 hours weekly; scope and duration vary by project.
- $60–80/hr is the observed band for this category — actual rates are set per project and are not guaranteed by applying.
- Applying joins a network. Matching is rolling, so there may be a gap between verification and your first invitation.
- You'll confirm work location during onboarding; some client projects have jurisdictional constraints.