For accountants

Accounting is one of the domains that looks simple to a language model until it is asked to be exactly right. A model can produce a plausible-looking journal entry, a plausible-looking depreciation schedule, or a plausible-looking tax opinion, and be confidently wrong in a way that would fail any working audit. Frontier AI labs know this. Their evaluation and training programs are increasingly staffed by practicing accountants for exactly this reason.

The work is precise. The demand is expanding. The pay is set against Western accounting-firm scales, in USD, on an hourly basis.

What the work looks like

• Response evaluation. You review model outputs on prompts covering financial reporting, audit reasoning, tax analysis, transfer pricing, revenue recognition, or accounting standards interpretation, and score them against what a working accountant would produce.
• Question authoring. You write realistic scenarios at the level of a case a working senior would see, dense enough for models to learn from.
• Standards checking. You verify specific claims against the applicable standards (IFRS, US GAAP, local GAAPs) and flag anywhere the model has produced a plausible-sounding but non-conforming answer.
• Adversarial testing. You construct prompts where the model is likely to miss a specific jurisdictional nuance, an edge case in a standard, or a substantive-over-form distinction.

The work is asynchronous. Two to four hour blocks. Deliverables are structured.

What labs actually look for

Professional credentials (CA, CPA, ACCA, or equivalent) or verifiable recent practice in a substantive area (audit, tax, corporate finance, controllership). Written English at the level a technical accounting memo expects.

They do not look for AI experience.

Where the pay lands

Rates are in USD. For accountants in emerging markets, this is typically well above local firm compensation on a per-hour basis. For accountants in higher-cost markets, it is competitive with consulting rates.

How to enter

Sign up, upload a recent CV that makes your credentials and current practice area visible, and let the platform surface open roles in accounting annotation, evaluation, and adversarial testing.