EU Regulatory Changes

This publication is a systematic academic review, not a regulatory change. It surveys the current state of federated learning techniques for detecting cyber intrusions in vehicular networks, such a...

This publication describes a novel cyberattack method targeting digital twins—virtual replicas of physical systems—using deep reinforcement learning. The attack is designed to degrade system compon...

This publication from arXiv, dated July 12, 2026, presents a research paper exploring whether watermarking techniques can serve as a technical safeguard against large language model (LLM) model ste...

This publication introduces a conceptual model for digital credential ecosystems that shifts the focus from credential issuers to verifiers, proposing a new framework for how credentials are valida...

This publication, "Distributed Denial of Science," identifies a novel risk under the AI Safety framework: indirect data poisoning of AI systems used in scientific research. The paper demonstrates h...

This paper, published on arXiv, introduces a novel technical framework called PromptGraph, which proposes a method for sanitizing user prompts sent to large language models (LLMs) to better balance...

This publication from arXiv presents a new technical method for detecting AI-generated images by analyzing patterns in image noise residuals. The proposed framework, called Noise Residual Clusterin...

This paper, published on arXiv, presents a new class of security vulnerability specific to multi-agent AI systems, termed "distributed backdoors." Unlike traditional backdoors triggered by a single...

This paper, published on arXiv, introduces a novel explainable AI system designed to detect conversational scams in real-time. The system uses a summary-based memory framework to track and analyze ...

This paper, published on arXiv, introduces a new automated framework called "Agent Hacks Agent" for red-teaming AI agents. Red-teaming is the practice of stress-testing systems to find vulnerabilit...