OpenAI has moved high-risk action constraints and sensitive data protection forward to the agent workflow and further to the platform layer

OpenAI dismantles proxy anti-prompt injection: high-risk actions begin to be pre-restrained, and sensitive data is protected into the workflow

OpenAI has published a technical article on how agents can resist prompt injection, and the core meaning is straightforward: the real danger is not re...