AI companies pay people to attack their own models: probing for jailbreaks, hunting failure modes and edge cases, stress-testing safety guardrails before the public finds the gaps. The work rewards a security mindset and creative persistence in equal measure. It is done from home, typically paid hourly, and flexible enough to run as a side gig alongside other work.
Applied Clinical Judgement tracks red-team and AI safety testing roles daily across Mercor, micro1 and Turing. We are a referral service, not the employer. Each card shows pay, hours and eligibility, and links straight to the platform’s application page.
Scroll down for 5 live AI Red Team & Safety Testing roles5 live AI Red Team & Safety Testing roles · updated daily
Red-Teaming Quality Assurance Lead (QAL)
$100 / hr. SME Careers seeks a Red-Teaming Quality Assurance Lead for remote contractor work overseeing safety-evaluation projects. You will review adversarial prompts, risk analyses, and trainer outputs; provide written feedback against detailed rubrics; identify quality gaps; and support remote expert teams via Discord. The role requires 3+ years in AI safety, red-teaming, or trust-and-safety work, strong English communication, and demonstrated expertise in evaluating model vulnerabilities and policy alignment. Note: no immediate project assignment, but qualified candidates join the expert network for future opportunities.
Note: SME Careers asks you to create a free account before it shows the role.
AI Jailbreak & Prompt-Injection Security Expert
$50-$90 / hr. micro1 seeks security specialists to evaluate AI system robustness through adversarial testing. You'll design methodologies for identifying jailbreaks, prompt injection vulnerabilities, and tool-use abuse scenarios. Work involves creating test frameworks, documenting findings, and collaborating with technical teams. A background in adversarial ML, red teaming, or AI safety is preferred, alongside demonstrated expertise—published research, open-source contributions, or conference participation. No prior AI experience required if you possess relevant domain knowledge.
Cyber Threat Intelligence SME
$70–$110/hr. micro1 seeks Cyber Threat Intelligence SMEs to evaluate AI models on advanced threat scenarios. You'll map CTI cases to taxonomies, author boundary cases on surveillance and attribution, grade threat responses, and identify over-refusal instances. Domain expertise in nation-state operations, APT tracking, and threat attribution is essential. Prior AI evaluation or red-teaming experience is valued. Remote contractor role.
Nuclear & Radiological Safety Expert for AI Model Audit
$90/hr. SME Careers seeks nuclear engineers, reactor physicists, health physicists, and safeguards specialists to audit AI model responses on sensitive nuclear and radiological topics. You will classify requests as benign, dual-use or adversarial, evaluate model safety, rank responses and draft ideal answers. The work suits professionals with five years' experience in nuclear fields, particularly those with licensing, inspection or red-teaming background.
Note: SME Careers asks you to create a free account before it shows the role.
AI Benchmark & Red-Teaming Specialist
Turing seeks an analytical researcher to author and test evidence-based web-research benchmark tasks that expose weaknesses in frontier AI models like Claude Opus 5 Max. You will discover obscure facts, construct precise research tasks, test them against state-of-the-art LLMs, and diagnose failure mechanisms in model reasoning and search strategies. The role requires 1+ years in investigative research, fact-checking, AI benchmarking, or red-teaming, plus intermediate-to-advanced OSINT skills and a bachelor's degree in a relevant engineering domain. Contractor engagement, up to 40 hours weekly with 4-hour PST overlap.
No live roles match your search.
AI training work is organised by profession, task and software — not by topic or sector. Try your field (for example “nursing” or “Python”), clear the filters, or browse the categories further down the page. The always-open talent pools below are a good place to start.
Frequently asked questions
Do I need a cybersecurity background?
It helps, but plenty of listings weigh adversarial creativity just as heavily: the knack for finding the prompt that breaks the rule. Conventional pen-testing skills transfer well. Each listing states its own requirements.
Is this legitimate work?
Yes. This is sanctioned testing commissioned by the AI companies on their own systems, under agreement, so weaknesses get fixed before deployment.
What does the day-to-day involve?
Writing adversarial prompts, documenting model failures systematically, and often rating responses from a safety perspective. Clear written English and methodical documentation matter as much as ingenuity.