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

Relevant Responsible AI Studio tools

More compliance + audit terms

See the full 80-term glossary →