Red-Teaming

A structured adversarial testing process in which a team attempts to find failures, vulnerabilities, and harmful outputs in an AI system. Borrowed from cybersecurity practice and adapted for AI safety evaluation. Required by the US Executive Order for developers of powerful AI systems and increasingly expected by regulators.