Red Teaming — структурированный процесс adversarial testing, используемый для выявления слабостей, unsafe behavior или harmful outputs в системе ИИ (AI). Он включает намеренное probing системы сложными, malicious или edge-case inputs, чтобы разработчики могли оценить и улучшить safeguards.
Red-teaming means a structured testing effort to find flaws and vulnerabilities in an AI system, often in a controlled environment and in collaboration with developers of AI. AI red-teaming is most often performed by dedicated 'red-teams' that adopt adversarial methods to identify flaws and vulnerabilities, such as harmful or discriminatory outputs from an AI system, unforeseen or undesirable system behaviors, limitations or potential risks associated with the misuse of the system.