Arabic Trust & Safety Data Trainer | $50/hr Remote

Arabic Trust & Safety Data Trainer Work Snapshot

• Job Type: Contract
• Location: Remote (France)
• Compensation: Up to $50 per hour
• Commitment: Flexible / project-based (hourly)
• Review and evaluate AI-generated responses for safety, accuracy, reasoning quality, and clarity
• Annotate and label content in Arabic and English to improve AI safety and policy adherence
• Assess step-by-step reasoning and identify unsafe, toxic, or policy-violating outputs
• Create and curate safety-focused training data, including adversarial and red‑team scenarios
• Compare and rank multiple model responses based on safety standards and quality benchmarks
• Identify failure patterns such as bias, misinformation, harmful content, or procedural risks
• Stress‑test AI systems to uncover policy gaps, edge cases, and inconsistencies
• Provide clear documentation and feedback to improve model reliability and safety

Requirements

• Education: Bachelor's degree or higher in a relevant field such as Communications, Linguistics, Psychology, Law/Policy, Security Studies, or equivalent experience
• Near-native or native proficiency in Arabic and strong English proficiency (C1 or above) with the ability to interpret policies, annotate content, and document decisions accurately
• Experience in Trust & Safety, content moderation, policy enforcement, risk operations, or safety evaluation, including proven LLM red‑team skills and familiarity with adversarial testing
• Strong knowledge of safety domains including hate and harassment, sexual content, self‑harm, violence, bias, illegal activities, malicious code, and misinformation
• Ability to handle exposure to sensitive and potentially disturbing content with emotional resilience and maintain consistent judgment under ambiguity
• Experience in AI data annotation, evaluation, or training workflows, along with familiarity using tools like Perplexity, Gemini, or ChatGPT for analysis and evaluation