Red Teaming
Adversarially probing AI for safety failures — getting it to produce harmful, biased, or dangerous outputs.
Definition
Red teaming is structured adversarial testing. Workers craft prompts designed to break safety guardrails (jailbreaks, edge cases, social-engineering attacks) and document where the model fails. It's a specialist niche; OpenAI, Anthropic, and frontier labs actively hire red teamers.
Examples
- Trying jailbreak techniques to get the model to produce restricted content
- Probing for racial or gender bias in completions
- Multi-turn social engineering of the model
Related terms
Testing AI outputs against quality criteria — the largest single category of AI training work.
Training AI to refuse harmful requests, avoid bias, and stay within policy.
Ensuring AI systems pursue goals that match human intent and values — the hardest open problem in AI.
Looking for actual gigs that use Red Teaming? Browse all live listings or filter by $100+/hr.