OpenAI updated its model guidelines: The first is 'Be True to Human Intent.'

OpenAI updated its model guidelines, with the first being 'Be True to Human Intent,' which requires models to go beyond literal obedience, judge users' true intentions by combining context, and emphasizes that memory usage must respect privacy and control. GPT is prohibited from self-enhancement, making money for OpenAI, must unconditionally cooperate with monitoring and shutdown, and immediate action outside the authorized scope must be stopped.