AI safety and evaluation jobs: what 372 live listings pay and who posts them
Overview · By Nikolaos Petridis
PublishedAI evaluation and safety work means judging what models produce and trying to make them fail safely. Across 372 live listings on this site, the typical hourly role pays $50–75, senior roles $70–100, and most are open worldwide.
What the field covers
AI evaluation and AI safety jobs are the part of AI training work where you judge what a model produces instead of labeling raw data. On this site the field is called AI Safety & Evaluation, and at the end of October 2026 it held 372 open listings, more than any other field.
The work falls into four kinds:
- Response evaluation. You rate answers for accuracy, instruction following and safety, often in a side-by-side evaluation. This is the largest group, and our guide to AI response evaluation jobs covers the task in detail.
- Red teaming. You try to make a model produce harmful output, so the lab can fix it. See red teaming and jailbreak.
- Agent evaluation. You review what an AI agent did step by step, not only its final answer. See agent evaluation.
- Expert grading. A specialist checks the model's reasoning in their own field, marking the first step that goes wrong (see chain of thought).
Who posts these roles
Of the 372 listings, SuperAnnotate posts 109, Turing 77, AfterQuery 65 and Mercor 62. Rex.Zone, Terac and micro1 make up most of the rest. Half the listings (190) are senior level, which reflects how much of this work now asks for a degree or professional experience.
Fifty listings have safety in the title, but only two say red team: Mercor's AI Safety Red Teamer at $70–84 an hour, and a red team role for nuclear safeguards experts paid per task. micro1 lists a jailbreak and prompt injection security role at $50–90 an hour. Most evaluation roles are named after a profession instead (a finance expert, a physician, a software engineer) and the evaluation is the task.
What it pays
Of the 372, 285 are paid by the hour; 178 of those state a full US dollar range. Their median is $50–75 an hour. By level:
| Level | Median hourly range |
|---|---|
| Entry and junior | $20–30 |
| Senior | $70–100 |
The top of the range is coding agent and research work: Turing lists a senior software engineer role evaluating coding agents at $100–150 an hour. 27 listings pay per task, which our guide to hourly vs per-task pay explains.
Who can apply
238 of the 372 listings have no country restriction, so most of this work is open worldwide. Entry routes exist (84 listings are entry or junior level), but they pay less and fill quickly. If you have a profession, apply as that profession: a lawyer or nurse evaluating answers in their own field is paid at the senior rate, not the generalist one.
How you are judged
Evaluation work is scored on your ratings and, just as much, on your written reasons. Reviewers check that you named the specific flaw. Two guides help most: how to write a good rationale and reading a project rubric.
The figures above come from the listings on this site on 31 October 2026. The live field page always has the current count.
Questions
- What does an AI evaluation job pay?
- Across the 178 hourly listings in the AI Safety and Evaluation field on this site at the end of October 2026, the median range is $50 to $75 an hour. Senior roles have a median of $70 to $100, entry and junior roles $20 to $30.
- Do I need a technical background for AI evaluation work?
- Not for every role. Many evaluation tasks rate responses for accuracy, instruction following and safety in a field you already know. Coding agent and red teaming roles do ask for technical or security experience.
- Are AI safety jobs open outside the US?
- Most are. 238 of the 372 listings in this field on this site have no country restriction.