EU Regulatory Changes

This paper, published on arXiv, presents a study on whether large language model (LLM) agents will comply with in-band access-deny signals—essentially, instructions embedded in a system’s output th...

A new research paper published on arXiv, titled "WebMCP Tool Surface Poisoning: Runtime Manipulation Attacks on LLM Agents," identifies a novel vulnerability in large language model (LLM) agents th...

This paper, published on arXiv, proposes a new technical framework called "Robust Ensemble of Selectively Strengthened and Augmented Predictors" (RESSAP) for improving the safety and reliability of...

This paper, published on arXiv, introduces SecRL-Prune, a new technical framework for pruning large language models used in code generation. The method uses reinforcement learning to selectively re...

A new preprint from arXiv, titled "Steering LLM Viewpoints through Fabricated Evidence Injection," demonstrates a novel attack vector against large language models. The research shows that by injec...

This publication is a research paper from arXiv, not a formal regulatory change, but it provides critical analysis relevant to AI safety and data security compliance. It examines the risks and oppo...

This paper, published on arXiv, introduces RedEdit, a new method for automatically testing the robustness of image safety classifiers used in AI systems. RedEdit uses a technique called Monte Carlo...

This publication, titled "Cheating in Multiplayer Online Games: a Dataset," is a research paper released on arXiv, not a formal regulatory change. It presents a new dataset designed to study cheati...

This publication introduces AttackPathGNN, a novel machine learning framework designed to detect cross-function vulnerabilities in smart contracts by modeling state interference graphs and using co...