Tags → #privacy risks

Research Identifies Security Vulnerabilities in Language Models

Recent research has revealed significant security vulnerabilities in large language models, particularly related to privacy risks in sensitive fields, due to timing-based side-channel attacks that exploit cache-sharing optimizations.
New Method for Membership Inference Attacks on Language Models

Researchers have developed a new method for Membership Inference Attacks (MIA) on large language models, called Self-calibrated Probabilistic Variation MIA (SPV-MIA), which aims to improve the accuracy of detecting whether specific data records were used in training.
Advancements in Membership Inference Attacks Research Introduced

A recent research paper presents the Loss Trace IQR (LT-IQR) method, which improves the efficiency of Membership Inference Attacks by identifying vulnerable data samples with reduced computational resources compared to traditional approaches.