What the work actually involves
You will spend most of your time producing Polish-language prompts across sensitive subject areas — the kind of material a model needs to handle carefully rather than refuse blindly — and then classifying prompts and conversations against a written guideline document. A typical session mixes authoring with review: drafting adversarial phrasings, spotting escalation patterns where a benign opening turns into a harmful request over several turns, and writing a short justification for every label you apply. The written rationale matters as much as the label; downstream teams use it to decide whether the guideline itself needs revising.
Cultural judgment is the reason the role is bilingual rather than translated. Polish idiom, euphemism, regional slang, and indirect phrasing all give attackers cover that an English-only reviewer would miss, and a phrasing that reads as harmless in literal translation can be plainly loaded to a native speaker. You should expect to explain, in English, why a Polish construction carries a meaning its surface words do not.
What the platform screens for
Mercor's screening is AI-led and conversational. Expect verification of Polish fluency at native or near-native level, evidence that your written English is clear enough for guideline-based rationales, confirmation of a bachelor's degree completed or in progress, and probing on judgment around dual-use information — where the line sits between a legitimate informational request and operational uplift. Prior trust and safety, moderation, or red-teaming experience is a plus rather than a gate, and follow-up questions tend to press for concrete examples rather than restatements of your résumé.
Logistics
- Remote and asynchronous; you choose your hours within task deadlines.
- Roughly 7 hours per week, part-time, with an immediate start.
- Observed pay band of $40–44/hr — stated as observed on the platform, not guaranteed.
- Poland or the wider Eastern Europe region is preferred, not required.