French AI Data Trainer - Remote

About the role

We are looking for a detail‑oriented French AI Data Trainer to join our remote team on an hourly contractor basis. You will help improve the safety and reliability of large language models by evaluating and annotating AI‑generated content in both French and English.

Key responsibilities

• Evaluate AI‑generated responses for accuracy, safety, and compliance with strict policy guidelines.
• Annotate and label safety‑critical datasets in French and English with high linguistic precision.
• Identify unsafe outputs related to hate, sexual content, violence, self‑harm, bias, misinformation, illegal activities, and malicious code.
• Compare multiple model outputs and score them using safety and quality rubrics.
• Detect model failure patterns such as evasion, escalation, or unsafe normalization.
• Conduct red‑teaming and adversarial testing to uncover vulnerabilities.
• Document findings clearly, explaining the reasoning behind each safety judgment.
• Support the refinement of annotation guidelines and edge‑case decision frameworks.
• Ensure consistent policy interpretation across all evaluations.

Required profile

• Bachelor’s degree or equivalent experience in linguistics, communications, psychology, law, policy, security studies, or a related field.
• Native or near‑native proficiency in French (reading and writing).
• Minimum C1 proficiency in English (reading and writing).
• Experience in trust & safety, content moderation, policy enforcement, risk operations, or safety evaluation.
• Proven experience red‑team­ing large language models and documenting findings.
• Strong analytical judgment and ability to work with ambiguous policy situations.
• High attention to detail and ability to explain decisions clearly and consistently.
• Comfortable working independently in a remote contractor setup.

Required skills

• ChatGPT
• Gemini
• Perplexity

What we offer

• Fully remote work environment.
• Hourly contract with flexible scheduling.
• Opportunity to contribute to cutting‑edge AI safety research.