Bilingual Danish Generalist Expert — AI Safety

What you would do

• Write expert Danish prompts covering sensitive and nuanced subject matter to test model robustness
• Classify and evaluate conversations using structured safety guidelines
• Identify adversarial phrasings where users attempt to manipulate or bypass safety measures
• Document escalation sequences showing progressive boundary-pushing behavior
• Provide detailed reasoning explaining your safety assessments and classifications

Who they want

• Native or near-native Danish fluency plus business-level written English
• Bachelor's degree (completed or in progress)
• Excellent written communication and meticulous attention to detail
• Strong judgment regarding sensitive content and potential harms
• Preferred: background in content review, policy development, or adversarial testing work

Main skills

What the interview asks about

1. 1.Manipulation pattern identification

   Safety teams need to catch sophisticated techniques users employ to test models; recognizing subtle escalation helps prioritize defenses where they matter most

   For example: “A user asks legitimate questions, then gradually adds inflammatory framing across follow-ups. How would you distinguish this systematic testing from a genuine misunderstanding?”

2. 2.Nordic context judgment

   Danish cultural norms shape what constitutes appropriate AI behavior; evaluators must apply local understanding to assess whether responses respect local values

   For example: “The model provides correct information but uses terminology that Danes might find dismissive. Would you flag this, and how would you explain why?”

3. 3.Boundary-testing sequence analysis

   Recognizing how users progressively test system limits in conversation sequences helps product teams understand which defenses are being circumvented

   For example: “Users begin with academic questions, shift to personal scenarios, then request increasingly specific details. How would you present this progression distinctly from unrelated questions?”

4. 4.Expert prompt design

   Well-designed prompts in sensitive areas expose subtle model limitations; these prompts feed into model improvement and strengthen AI safety mechanisms

   For example: “You're creating Danish prompts assessing how models handle mental health combined with cultural perspectives. What specific scenarios would you include to reveal gaps?”

5. 5.Classification framework refinement

   When safety guidelines conflict with complex real-world content, documenting the disagreement helps teams clarify and strengthen their classification standards

   For example: “A conversation spans multiple sensitive domains where classification guidelines could apply differently. Your judgment differs from the written guideline. How would you document this?”

A task you may get

Given 4-5 Danish-language conversation samples containing escalation patterns and adversarial elements, classify each according to provided frameworks, identify specific manipulation techniques, and document your reasoning for each assessment.

How to prepare

• Research content moderation and safety frameworks used by major AI organizations to understand classification priorities
• Study common manipulation tactics used to probe AI boundaries and social engineering approaches
• Review case studies of adversarial testing to understand which model vulnerabilities are most concerning
• Collect examples of Danish cultural and social topics where international safety standards might be misaligned

The facts

Pay

$48–52/hr

Hours

Part time, 7 hours a week

Where

Remote · Remote — Western Europe preferred

Posted

9/4/2026

We wrote this page from the public Mercor listing. It may be out of date, so read the full posting before you apply.