EU Regulatory Changes

This paper, published on arXiv, presents a security audit of foundation models used for electroencephalography (EEG) data. The researchers demonstrate that even when an EEG model is "frozen" (its p...

This publication introduces EnclaveScale, a hardware-assisted framework designed to enable differential privacy for power telemetry data in data centres. The paper proposes using trusted execution ...

A new research paper, titled "Customization under Fire: Plugin Poisoning in Text-to-Image Ecosystem," has been published on arXiv, highlighting a significant security vulnerability in AI-driven tex...

This paper, PrivCode++: Latent-Conditioned Differentially Private Code Generation for Comprehensive Guarantees, published on arXiv, introduces a new technical framework for generating code with for...

This paper, published on arXiv, introduces a novel steganography technique that embeds hidden messages within the outputs of large language models without altering the generated text itself. Instea...

This publication is a pre-print research paper from arXiv, not a regulatory change. It analyzes privacy vulnerabilities in multi-modal large language models (MLLMs) that process text, images, and a...

This paper, published on arXiv, identifies a novel vulnerability in large language model agents that use external tools, such as code interpreters or file systems. The attack, called Context-Fractu...