Contextual Integrity in LLMs via Reasoning and Reinforcement Learning

TL;DR

This paper introduces a reasoning and reinforcement learning approach to improve contextual integrity in large language models, reducing inappropriate information sharing while maintaining task performance, and transferring well to real-world benchmarks.

Contribution

It presents a novel RL framework that enhances LLMs' reasoning about context and privacy norms, improving information disclosure control.

Findings

Significant reduction in inappropriate disclosures.

Maintains task performance across models.

Transfers improvements to real-world benchmarks.

Abstract

As the era of autonomous agents making decisions on behalf of users unfolds, ensuring contextual integrity (CI) -- what is the appropriate information to share while carrying out a certain task -- becomes a central question to the field. We posit that CI demands a form of reasoning where the agent needs to reason about the context in which it is operating. To test this, we first prompt LLMs to reason explicitly about CI when deciding what information to disclose. We then extend this approach by developing a reinforcement learning (RL) framework that further instills in models the reasoning necessary to achieve CI. Using a synthetic, automatically created, dataset of only $\sim700$ examples but with diverse contexts and information disclosure norms, we show that our method substantially reduces inappropriate information disclosure while maintaining task performance across multiple model sizes and families. Importantly, improvements transfer from this synthetic dataset to established CI benchmarks such as PrivacyLens that has human annotations and evaluates privacy leakage of AI assistants in actions and tool calls. Our code is available at: https://github.com/EricGLan/CI-RL