Can Language Models be Instructed to Protect Personal Information?

TL;DR

This paper introduces PrivQA, a benchmark for evaluating privacy-utility trade-offs in multimodal language models, and proposes a self-moderation technique to enhance privacy, while also revealing vulnerabilities to simple adversarial attacks.

Contribution

It presents PrivQA as a new benchmark for assessing privacy protections and introduces a self-moderation method to improve privacy in multimodal models.

Findings

Self-moderation significantly improves privacy protection.

Adversaries can bypass protections with simple jailbreaking methods.

PrivQA can aid in developing more robust privacy-preserving models.

Abstract

Large multimodal language models have proven transformative in numerous applications. However, these models have been shown to memorize and leak pre-training data, raising serious user privacy and information security concerns. While data leaks should be prevented, it is also crucial to examine the trade-off between the privacy protection and model utility of proposed approaches. In this paper, we introduce PrivQA -- a multimodal benchmark to assess this privacy/utility trade-off when a model is instructed to protect specific categories of personal information in a simulated scenario. We also propose a technique to iteratively self-moderate responses, which significantly improves privacy. However, through a series of red-teaming experiments, we find that adversaries can also easily circumvent these protections with simple jailbreaking methods through textual and/or image inputs. We believe PrivQA has the potential to support the development of new models with improved privacy protections, as well as the adversarial robustness of these protections. We release the entire PrivQA dataset at https://llm-access-control.github.io/.