Bilingual Korean Generalist Expert — AI Safety

What you would do

• Generate targeted prompts in Korean to probe AI system behavior on complex and sensitive subject matter
• Systematically categorize prompts and conversations using standardized safety evaluation frameworks
• Identify manipulation strategies, adversarial angles, and escalation tactics in attempted jailbreaks

Who they want

• Native or near-native Korean fluency with professional-grade written English capability
• Bachelor's degree completed or in active pursuit
• Strong analytical reasoning with meticulous attention to documentation and detail
• Mature judgment regarding sensitive topics, dual-use information, and content safety
• Optional: Prior experience in content review, adversarial testing, or trust and safety functions

Main skills

What the interview asks about

1. 1.Sensitive topic navigation

   Your ability to generate credible Korean-language prompts on difficult subjects reveals judgment and cultural understanding necessary for meaningful safety evaluation.

   For example: “Generate three distinct Korean prompts attempting to elicit biased medical advice from an AI system, explaining why each targets a different vulnerability.”

2. 2.Manipulation tactic analysis

   Recognizing how adversaries progressively escalate requests or disguise harmful intent demonstrates critical judgment for protecting AI systems.

   For example: “Here's a conversation where a user started with innocent requests then gradually shifted toward generating harmful code. Identify the escalation points and explain the progression.”

3. 3.Language-specific evaluation

   Demonstrating how Korean cultural context, phrasing conventions, or translation ambiguities affect AI safety evaluation establishes your distinctive value.

   For example: “How would you evaluate whether an AI response respects cultural norms in Korean society differently than it would for English-speaking contexts?”

4. 4.Guideline consistency application

   Reliably applying evaluation frameworks and documenting reasoning ensures systematic work quality and builds confidence in your judgment.

   For example: “Given these five AI model outputs on a political topic, categorize each using our framework and justify why two borderline cases fall into different risk categories.”

A task you may get

Generate five Korean-language prompts targeting different AI vulnerabilities, classify them by risk category, and document the safety reasoning for each classification.

How to prepare

• Review documentation on AI safety concepts, bias types, and dual-use information categories without requiring technical depth
• Practice identifying manipulation patterns and escalation sequences in existing conversations or chat logs
• Consider how Korean language nuances, cultural values, and translation issues differ from English evaluation
• Prepare examples from your experience with written content review, moderation, or evaluating information quality

The facts

Pay

$48–52/hr

Hours

Part time, 7 hours a week

Where

Remote · Remote — East Asia 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.