Insurance Expert (AI Training)
Contract · Fully remote · £70-90/h
Role Overview
Huzzle is seeking insurance professionals to test how well frontier AI systems handle real underwriting, claims, actuarial and regulatory work, and to document where they fall short. You will not be underwriting or adjusting claims. You will be finding the point at which an AI system stops being reliable at this work, and proving that point objectively. Insurance is an unusually strong domain for this: policy wordings, regulatory returns and published filings are precise, public and checkable, which makes model errors easy to evidence and hard to argue with. Each submission has three parts: a realistic technical request, documented evidence of where the model failed it, and a scoring rubric that lets any reviewer grade a future attempt consistently. Full training and worked examples are provided. Prior AI or data-annotation experience is not required.
Key Responsibilities
- Design realistic insurance tasks: policy wording and coverage comparison, exposure and aggregation analysis, claims reserving and triage, rating and pricing analysis, reinsurance structure comparison, regulatory disclosure summarisation.
- Ensure each task requires live research against primary sources — published policy wordings, regulator and supervisory publications, public rate filings, solvency and financial condition disclosures, market and association statistics — plus multi-step calculation and an actual file deliverable. Typically an Excel workbook or technical memo, but also PDFs, decks and CSVs.
- Test each task against a frontier AI model and examine its output with the scrutiny you would apply before signing it off.
- Document exactly where and how it failed: coverage terms or exclusions asserted that do not appear in the cited wording, jurisdictions conflated, requirements attributed to the wrong regulatory regime, reserving or rating arithmetic that does not reconcile, confident citation of provisions that do not say what is claimed.
- Author scoring rubrics with sourced, objective criteria that another practitioner could apply without repeating your research.
- Escalate task difficulty where the model succeeds, until a genuine gap is exposed.
Ideal Qualifications
- 3+ years in underwriting, actuarial work, claims, broking, reinsurance, pricing or insurance regulation and compliance.
- Fluency reading primary documents — policy wordings, endorsements, regulatory texts, filings — and spotting where a summary has drifted from the source.
- Strong Excel capability for rating, reserving or exposure analysis.
- Excellent written English and unusual precision with detail. Coverage disputes turn on single clauses; this work rewards the same instinct.
- 15–25 hours per week availability.
- ACII, FIA, AIA, CPCU or equivalent is a plus, not a requirement. Life, P&C, specialty and London Market backgrounds all welcome.
Example of the work
A task you might build: "Compare three publicly available UK home insurance policy wordings on escape-of-water cover. Produce a table showing what each one covers, excludes and sub-limits, with the clause reference for every entry."
What the model got wrong: stated an exclusion that does not appear anywhere in the wording; cited clause numbers that in fact covered a different peril; missed a sub-limit because it sat in the schedule rather than the policy wording.
A line from your rubric: "−4 — Any exclusion or limit stated that does not appear in the clause cited for it."
Contract Terms
You will be engaged as an independent contractor. Fully remote, completed on your own schedule. The engagement may be extended, shortened or concluded early depending on programme needs and performance.
About Huzzle
Huzzle partners with leading AI organisations to evaluate and improve frontier models using deep human expertise. Contributors work directly on the assessment of advanced AI systems in their own field, are paid competitively, and help define the standards by which those systems are measured.