OpenAI's update injects OpenAI's public proxy anti-prompt into the design

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...