Red Teaming

Deliberately testing an AI system by attempting to make it produce harmful, incorrect, or unintended outputs. Red teams probe for safety vulnerabilities, bias, and failure modes. It is a critical part of responsible AI development before deployment.