Glossary · Compliance + audit
Red Teaming
Structured adversarial evaluation of an AI system — testers deliberately try to elicit harmful, biased, or policy-violating outputs. Standard practice for foundation-model providers pre-release. NIST AI 600-1 recommends red teaming for generative AI; EU AI Act Article 55 effectively requires it for GPAI models with systemic risk.
Framework references
- NIST AI 600-1
- EU AI Act Art. 55