AIBearisharXiv – CS AI · Jun 237/10
🧠Researchers introduced MuPPET, a benchmark testing privacy vulnerabilities in large language model assistants operating in multi-party conversations. The study reveals that LLMs leak significantly more sensitive information in group settings than in one-to-one interactions, with both frontier and smaller open-weight models showing substantial exposure risks that existing privacy defenses cannot adequately address.
AIBearisharXiv – CS AI · Jun 107/10
🧠Researchers introduced IDP-Bench, the first benchmark evaluating how well large language models protect interdependent privacy—where one person's data can be revealed by others without consent. Testing eight open-source LLMs revealed strong performance in recognizing data co-ownership but significant weaknesses in understanding contextual integrity parameters and judging sharing appropriateness, with smaller models showing particular vulnerability to prompt sensitivity.
AIBullisharXiv – CS AI · Apr 107/10
🧠Researchers introduce ConfusionPrompt, a privacy framework for large language models that decomposes user prompts into smaller sub-prompts mixed with pseudo-prompts before sending to cloud servers. The method protects user privacy while maintaining higher utility than existing perturbation-based approaches and works with existing black-box LLMs without modification.
AINeutralarXiv – CS AI · Jun 236/10
🧠Researchers introduce π-RAG, a novel retrieval architecture that protects sensitive data in Large Language Models by using the digits of pi as an oblivious indirection layer, eliminating direct exposure of vector embeddings to inversion attacks. The system combines semantic quantization with cryptographic salting to enable privacy-preserving retrieval for compliance-heavy sectors like finance and healthcare.
AIBullisharXiv – CS AI · Jun 96/10
🧠FuseFSS is a new compiler that streamlines secure LLM inference by consolidating fragmented protocol designs into a unified pipeline, achieving 1.24x-1.50x speedup and reducing communication overhead by 9-16% compared to existing function secret sharing approaches. The technology enables privacy-preserving queries to large language models without revealing user prompts, addressing a critical bottleneck in cryptographic systems for AI inference.
AINeutralarXiv – CS AI · Jun 46/10
🧠Researchers introduce DelegateCI-Bench, a privacy-focused benchmark for query rewriting in LLM delegation, combined with a reinforcement learning framework that selectively redacts sensitive information while preserving task-critical content. The approach achieves superior privacy-utility tradeoffs compared to existing type-based PII redaction methods, addressing growing concerns about sensitive data exposure in cloud-hosted AI systems.
AINeutralarXiv – CS AI · May 116/10
🧠Researchers introduce SHRED, a machine unlearning method for large language models that removes memorized private or copyrighted data without requiring a curated retain set of examples. By selectively demoting logits of high-information tokens while preserving model utility through self-distillation, SHRED achieves superior trade-offs between forgetting efficacy and performance compared to existing retain-set-dependent approaches.
AINeutralarXiv – CS AI · Apr 156/10
🧠Researchers introduce a sequential unlearning framework that enables Large Language Models to forget sensitive data while maintaining performance, addressing GDPR compliance and the Right to be Forgotten in politically sensitive deployments. The method stabilizes general capabilities through positive fine-tuning before selectively suppressing designated patterns, demonstrating effectiveness on the SemEval-2025 benchmark with minimal accuracy degradation.
AINeutralarXiv – CS AI · Apr 106/10
🧠Researchers formalize privacy-preserving communication for LLM agents by introducing Information Sufficiency (IS) as a framework and proposing free-text pseudonymization as a third privacy strategy alongside suppression and generalization. Evaluation across 792 scenarios reveals that pseudonymization offers superior privacy-utility tradeoffs, and that multi-turn conversational testing exposes significant privacy leakage missed by single-message assessments.
