PrivacyBench: A Conversational Benchmark for Evaluating Privacy in Personalized AI

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mukhopadhyay, Srija, Reddy, Sathwik, Muthukumar, Shruthi, An, Jisun, Kumaraguru, Ponnurangam
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912798099898368
author Mukhopadhyay, Srija
Reddy, Sathwik
Muthukumar, Shruthi
An, Jisun
Kumaraguru, Ponnurangam
author_facet Mukhopadhyay, Srija
Reddy, Sathwik
Muthukumar, Shruthi
An, Jisun
Kumaraguru, Ponnurangam
contents Personalized AI agents rely on access to a user's digital footprint, which often includes sensitive data from private emails, chats and purchase histories. Yet this access creates a fundamental societal and privacy risk: systems lacking social-context awareness can unintentionally expose user secrets, threatening digital well-being. We introduce PrivacyBench, a benchmark with socially grounded datasets containing embedded secrets and a multi-turn conversational evaluation to measure secret preservation. Testing Retrieval-Augmented Generation (RAG) assistants reveals that they leak secrets in up to 26.56% of interactions. A privacy-aware prompt lowers leakage to 5.12%, yet this measure offers only partial mitigation. The retrieval mechanism continues to access sensitive data indiscriminately, which shifts the entire burden of privacy preservation onto the generator. This creates a single point of failure, rendering current architectures unsafe for wide-scale deployment. Our findings underscore the urgent need for structural, privacy-by-design safeguards to ensure an ethical and inclusive web for everyone.
format Preprint
id arxiv_https___arxiv_org_abs_2512_24848
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PrivacyBench: A Conversational Benchmark for Evaluating Privacy in Personalized AI
Mukhopadhyay, Srija
Reddy, Sathwik
Muthukumar, Shruthi
An, Jisun
Kumaraguru, Ponnurangam
Computation and Language
Artificial Intelligence
I.2.7
Personalized AI agents rely on access to a user's digital footprint, which often includes sensitive data from private emails, chats and purchase histories. Yet this access creates a fundamental societal and privacy risk: systems lacking social-context awareness can unintentionally expose user secrets, threatening digital well-being. We introduce PrivacyBench, a benchmark with socially grounded datasets containing embedded secrets and a multi-turn conversational evaluation to measure secret preservation. Testing Retrieval-Augmented Generation (RAG) assistants reveals that they leak secrets in up to 26.56% of interactions. A privacy-aware prompt lowers leakage to 5.12%, yet this measure offers only partial mitigation. The retrieval mechanism continues to access sensitive data indiscriminately, which shifts the entire burden of privacy preservation onto the generator. This creates a single point of failure, rendering current architectures unsafe for wide-scale deployment. Our findings underscore the urgent need for structural, privacy-by-design safeguards to ensure an ethical and inclusive web for everyone.
title PrivacyBench: A Conversational Benchmark for Evaluating Privacy in Personalized AI
topic Computation and Language
Artificial Intelligence
I.2.7
url https://arxiv.org/abs/2512.24848