Red Teaming

Red teaming is adversarial testing of LLM systems. Testers try to break system prompts, extract private data, cause harmful outputs, trigger hallucinations, and find edge cases the developers missed. Automated red teaming uses one LLM to attack another at scale. Frameworks like Garak, PyRIT, and custom harnesses make this systematic.