Organizations increasingly use LLMs to retrieve internal documents, but deciding who may see a given document is a privacy problem. The new arXiv paper "Privacy Conductor" argues that existing approaches based on Contextual Integrity ask the LLM itself to make that decision, meaning access depends on the model's judgment. The authors propose an architectural enforcement instead, separating the access decision from the retrieval model.
Because the abstract is truncated, the full mechanism is not described in the source. What is clear is the paper's central claim: privacy should be enforced by the system architecture, not left to the LLM. This aligns with a broader trend toward externalizing safety and privacy controls rather than relying on model behavior.