Reducing Privacy leaks in AI: Two approaches to contextual integrity

New research explores two ways to give AI agents stronger privacy safeguards grounded in contextual integrity. One adds lightweight, inference-time checks; the other builds contextual awareness directly into models through reasoning and RL. The post Reducing Privacy leaks in AI: Two approaches to contextual integrity  appeared first on Microsoft Research.