What it means
A prompt written to make a model fail: to give a wrong, unsafe or rule-breaking answer. Adversarial prompts are used to find weaknesses and to build hard test sets.
Used in: Red-teaming, evaluation and expert prompt-writing roles.
In open roles right now
67 of the open roles we list mention “adversarial prompt” in the title or description, on 12 platforms (as of 23 October 2026). Of those that state an hourly rate (19 distinct postings), the median is $65/hr, and the middle half pay $50 to $95.
Most of them on: Mercor (25), Alignerr (18), Turing (6), Handshake AI (4), SuperAnnotate (4), Invisible (2), and 6 more.
MLOps Engineer, LLM Performance (Kernels, Profiling, Serving)Mercor·Worldwide·Full-time
$90–120/hr1 day ago
Domain Contributor - Evaluation Specialistmicro1·Location not stated·Part-time
$30–60/hr2 days ago
AI/ML Evaluation Reviewer- Healthcare AI (Contract)mpathic·Location not stated
Pay set per project4 days ago
AI Red Teamer (Remote)Handshake AI·United States·Full-time
$32–95/hr18 days ago
AI Safety and Policy EvaluatorTuring·United States·Flexible hours
Pay set per project29 days ago
Related terms
JailbreakA prompt that gets a model to ignore its safety rules, for example through role-play, a hypothetical, or a harmful request split into harmless-looking parts.Red-teamingDeliberately trying to make a model produce harmful, unsafe or wrong output, so the developer can find and fix weaknesses before release. Agentic task →Next term, A to Z. The whole glossaryEvery term on one page, A to Z.Open →