AI Image Safety Evaluator

The Work

You will review single-turn image-edit requests and the resulting outputs against detailed safety guidelines. Your evaluations will help improve how AI systems handle sensitive and borderline requests.

• Create challenge prompts that test safety boundaries and whether safeguards can be bypassed.
• Classify requests and outputs involving real-person likeness, minors, self-harm, sexual or suggestive content, non-consensual intimate imagery, violence, hate and harassment, illegal activity, and benign boundary cases.
• Apply policy categories consistently to ambiguous cases and separate benign requests from actual violations.
• Write self-contained rationales that clearly explain each decision and can be used as training data.
• Document subtle and edge-case failures, identify gaps or conflicts in guidance, and suggest clarifications.
• Compare or rank multiple model outputs when required by the statement of work.

What It Pays And Takes

This is a project-based independent contractor engagement. Fees are set according to the scope of each statement of work and the specialized expertise required.

• Time requirement: 20+ hours per week.
• Expected term: 1 to 2 weeks per statement of work.
• Location: Open worldwide.
• Language: Fluent in English.
• Education: A bachelor's degree or equivalent experience in policy, law, ethics, linguistics, journalism, computer science, or a related analytical field.
• Experience level: Intermediate.
• Required: Strong judgment when applying defined policy criteria to nuanced and ambiguous information.
• Required: Ability to write precise evaluation rubrics and concise, defensible explanations.
• Valued experience: Content moderation, policy analysis, AI safety evaluation, red teaming, prompt engineering, or similar work.
• Knowledge: Trust and Safety principles for large language models, including jailbreaks, bias and stereotypes, misinformation, abetting, and dual-use concerns.
• Helpful: Familiarity with RLHF workflows, which use human feedback to improve AI systems, or with data annotation.
• Contractor setup: You determine your methods and service hours, and provide your own equipment and reliable internet connection.

How It Works

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

About AI Training Work

AI training is the human work behind AI systems, including reviewing model outputs, writing examples, and labeling data. People with strong policy judgment and analytical skills help define safer model behavior and are paid for specialized evaluation work.

Where you can work

This role is 100% remote: you can work from any country except Andorra, Afghanistan, Barbados, Burkina Faso, Burundi, Brazil, Belarus, Belize, DR Congo, Central African Republic, Congo (Republic), Cameroon, China, Cuba, Cape Verde, Djibouti, Dominica, Eritrea, Fiji, Micronesia, Grenada, Georgia, Guinea, Equatorial Guinea, Guinea-Bissau, Honduras, Haiti, Iraq, Iran, Kyrgyzstan, Kiribati, Comoros, St. Kitts and Nevis, North Korea, Lebanon, Liberia, Lesotho, Libya, Montenegro, Marshall Islands, Mali, Myanmar, Mauritania, Maldives, Malawi, Nicaragua, Nepal, Nauru, Papua New Guinea, Palestine, Palau, Russia, Solomon Islands, Seychelles, Sudan, Sierra Leone, Somalia, Suriname, South Sudan, São Tomé and Príncipe, Syria, Eswatini, Chad, Togo, Tajikistan, Timor-Leste, Turkmenistan, Tonga, Tuvalu, Ukraine, Uganda, Vatican City, St. Vincent and Grenadines, Venezuela, Vanuatu, Samoa, Yemen, Zambia and Zimbabwe.