EU Regulatory Changes

This paper, published on arXiv, introduces a new technical framework for learning the structure of causal relationships within REST APIs, specifically designed to support AI safety compliance. It p...

This paper, published on arXiv, proposes a new technical framework called "steerability via constraints" for improving the oversight of AI coding agents. It does not represent a binding regulatory ...

This publication, "Cloak and Detonate: Scanner Evasion and Dynamic Detection of Agent Skill Malware," presents new research demonstrating how advanced AI-driven malware can evade current static sec...

This paper, published on arXiv, presents a new framework called AI_SAFETY, which proposes a structured approach to preventing catastrophic failures in AI systems that control physical machinery, su...

This paper, published on arXiv, provides a comprehensive technical analysis of risk assessment and management methodologies for intelligent systems under the EU AI Act. It does not represent a new ...

This publication introduces a new technical framework for privacy-preserving and verifiable approximate distributed coded computing, which addresses how large-scale data processing tasks can be sec...

This publication, released on arXiv in July 2026, presents a technical paper titled "Behind the Refusal: Determining Guardrail Activation via Behavioral Monitoring." It does not represent a new reg...

A new technical paper, HaloGuard 1.0, has been published on arXiv, introducing an open-weights constitutional classifier designed to enhance multilingual AI safety. This is not a regulatory change ...

A new preprint, arXiv: kNNGuard, proposes a training-free, configurable guardrail for large language models (LLMs) that works by analyzing the model's internal hidden activations rather than relyin...