What the work involves

A frontier AI lab is training a model to produce professional presentation material, and risk advisory decks are among the hardest cases: dense, hierarchical, audience-calibrated, and often written to survive regulator scrutiny. You will work in three modes. First, authoring reference decks — a Q3 remediation status pack for an audit committee, a control framework exhibit mapping COSO components to testing coverage, a regulatory readiness roadmap with workstreams and gating milestones. Second, evaluating model output against a rubric, marking where the structure, the numbers, the hedging, or the visual hierarchy fails. Third, rewriting weak output into something you would actually put in front of a CRO, with a written rationale for each change.

The distinction that matters most here is between a deck that looks plausible and a deck that would survive a board risk committee. Models are good at producing a slide titled "Key Risks" with four bullets. They are bad at knowing that a board audience needs the residual risk position before the methodology, that an overdue remediation item cannot be described in the same register as an on-track one, and that an exhibit implying assurance the testing never provided is a real problem rather than a stylistic one.

What the platform screens for

  • Verifiable practice background. Four-plus years in risk, regulatory, or internal audit advisory at a top firm — Big 4 or equivalent. Expect follow-ups on the specific engagements, sectors, and regulatory regimes you worked under.
  • Genuine board- and regulator-facing exposure. Having drafted material that went into a board pack is different from having sat in a team that produced one. Screens probe for what you personally wrote and what came back in review.
  • Slide craft as a technical skill. Not decoration — structure, action titles, exhibit logic, what belongs on the page versus in the appendix, how a dense control matrix is made legible.
  • Evaluation judgment. Whether you can articulate why a deck fails, in terms another reviewer could apply consistently, rather than saying it feels off.

Logistics

Fully remote and asynchronous, with no fixed hours. Most contributors run 5–20 hours per week; higher volume is generally available for people whose output holds up in quality review. The $80/hour figure is what has been observed on this listing, not a guarantee — rates on these programmes move with task type and reviewer tier. The interview process includes an AI voice screen, which is a real conversation with follow-up questions, not a form. Confidentiality obligations from prior employers apply; you will be expected to build illustrative material, not to reproduce client work.