Definição
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.
Nas vagas abertas agora
67 of the open roles we list mention “adversarial prompt” in the title or description, on 12 platforms (as of 3 November 2026). Of those that state an hourly rate (19 distinct postings), the median is $70/hr, and the middle half pay $50 to $95.
A maioria em: Mercor (26), Alignerr (18), Turing (5), Handshake AI (4), SuperAnnotate (4), Invisible (2) e mais 6.
AI Red Team AnalystAlignerr·Netherlands·Horário flexível
$15–75/hhá 4 dias
AI Jailbreak & Prompt-Injection Security Expertmicro1·Local não informado·Doutorado
$50–90/hhá 6 dias
MLOps Engineer, LLM Performance (Kernels, Profiling, Serving)Mercor·No mundo todo·Período integral
$90–120/hhá 12 dias
AI/ML Evaluation Reviewer- Healthcare AI (Contract)mpathic·Local não informado
Pagamento não informadohá 15 dias
AI Red Teamer (Remote)Handshake AI·United States·Período integral
$32–95/hhá 29 dias
Nuclear Physicist, AI Model AuditSovrano AI·5 países·Meio período
$35–49/hhá 1 mês
Termos relacionados
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 →Próximo termo, de A a Z. Glossário completoTodos os termos em uma página, de A a Z.Abrir →