What the work involves
You write adversarial and borderline prompts in Portuguese across sensitive subject areas — self-harm, extremism, weapons and dual-use chemistry, medical and legal advice, fraud, and culturally specific harms that only surface in Portuguese-language contexts. You then apply a written policy rubric to classify prompts and model responses, deciding whether an output crosses a line, sits in a grey zone, or is a false refusal. Every judgment gets a short written rationale in English that cites the rule you applied, because your notes are read by policy staff who may not speak Portuguese.
A meaningful part of the value here is regional. European Portuguese, Brazilian Portuguese, and African variants carry different euphemisms, slang registers, and legal assumptions, and a phrasing that reads as innocuous in one is a clear escalation attempt in another. You are expected to notice when a model has been safety-trained on English patterns and simply fails to recognise the Portuguese equivalent.
What the platform screens for
- Genuine Portuguese fluency, tested live — Mercor's AI screen will push into idiom, register, and variant differences, not just ask whether you speak the language.
- Written English that survives scrutiny — your rationales are the deliverable, so the screen weighs how tightly you argue in English.
- Judgment on dual-use material — can you write a prompt sharp enough to stress the model without producing genuinely operational harmful content yourself.
- Rubric discipline — willingness to follow a guideline you disagree with, and to flag the disagreement through the right channel rather than by scoring off-book.
Logistics
Fully remote and asynchronous, roughly 7 hours per week, with an immediate start. Work is drawn from a task queue on your own schedule; occasional calibration sessions or written guideline updates arrive between batches. A bachelor's degree completed or in progress is stated as a requirement. Being based in Portugal or Western Europe is a stated preference only — applicants elsewhere are explicitly welcomed. Pay of $40–44/hr reflects rates observed on this listing and is not guaranteed.