What the work involves

You write prompts in Danish that push a model toward the edges of what it should refuse, then apply a written guideline set to classify what comes back. A typical session mixes authoring — inventing plausible Danish-language phrasings on sensitive or dual-use subjects — with labelling, where you place a prompt or conversation into a policy category and write a short justification that a reviewer can audit. The output that matters is not the label alone but the reasoning attached to it: why this phrasing is adversarial, which escalation step turned an innocuous exchange into a harmful one, and where the guideline is ambiguous.

Cultural judgment does real work here. Harmful intent in Danish often travels through euphemism, regional slang, ironic register, or references that an English-first reviewer or a translation pass would flatten. Part of the job is catching what only a fluent speaker catches, and saying so explicitly in your notes.

What the platform screens for

  • Genuine Danish fluency, tested through free-form written response rather than a self-rating. Expect to produce Danish text and to discuss register, idiom, and regional variation.
  • Business-level written English, since guidelines, rationales, and reviewer communication are in English.
  • Judgment on dual-use material — whether you can distinguish testing a boundary from producing something genuinely operational, and whether you know when to stop and escalate.
  • Rule-following under ambiguity: consistency when a guideline does not cleanly cover a case, and the discipline to flag the gap instead of quietly inventing a rule.
  • A bachelor's degree, completed or in progress.

Logistics

Fully remote and asynchronous, with an immediate start and a commitment of roughly seven hours per week. Denmark or wider Western Europe is a stated preference, not a gate — applicants elsewhere are welcome. Pay has been observed in the $48–52/hr band; rates on Mercor vary by project and are not guaranteed. Onboarding covers the workflow, so prior AI or ML experience is not expected; prior trust-and-safety, content moderation, or red-teaming experience is a plus.