AI Trainer Data Annotation Specialist (Safety)

You will support AI safety policy evaluation by reviewing AI-generated conversations, source materials, and model outputs. You will use project rubrics to make consistent judgments and explain the evidence behind them.

Some content may involve self-harm, violence, and abuse. The work requires careful reading, objective writing, and attention to subtle differences in meaning.

• Evaluate model outputs against detailed AI safety policies and project guidelines.
• Apply rating rubrics and document evidence-based judgments.
• Write clear, concise feedback that can help improve model behavior.
• Maintain accuracy and consistency across repeated review tasks.
• Review sensitive content involving self-harm, violence, and abuse.

This is part-time remote contractor work. The listing does not provide a compensation amount.

• Pay: Compensation details are not provided in the listing.
• Schedule: About 20 hours per week for an estimated 1 to 3 months, with flexible scheduling.
• Location: Open worldwide.
• Language: Excellent English proficiency is required.
• Experience: Entry-level; prior AI training experience is not required.
• Requirements: Excellent reading comprehension, close-reading ability, clear and objective writing, strong attention to detail, sound judgment, and the ability to follow detailed safety policies.
• Work style: You must be able to work accurately under time constraints.
• Knowledge: Familiarity with US-specific references, systems, and terminology is required.
• Helpful background: Data annotation, content moderation, policy review, trust and safety, AI training, response evaluation, editorial review, proofreading, compliance, quality assurance, structured rating systems, or evaluation rubrics.

Apply on OpenTrain with your resume, then complete the application on the hiring site.

AI training is the human work behind systems that learn from examples, including reviewing model responses and rating them against clear standards. People with strong subject knowledge and careful judgment help make those systems more accurate, useful, and safe.