EU Regulatory Changes

A new academic paper published on arXiv, titled "Blind PRNG Hijacking: An Undetectable Integrity-Preserving Attack Against LLM Watermarking," presents a novel method to remove or bypass watermarkin...

A new position paper published on arXiv, titled "Retire the 'Positive Backdoor' Label -- Secret Alignment Requires Strict and Systematic Evaluation," argues that the AI safety community should aban...

This technical report, published on arXiv on May 27, 2026, identifies emerging security and safety risks within the rapidly growing ecosystem of AI agent skills—modular capabilities that can be dow...

This paper, published on arXiv, introduces a novel method for detecting and exploiting refusal signals in large language models (LLMs) by analyzing their internal activations before a final output ...

This publication introduces the ADARE framework, which applies test-driven forensics to desktop investigations. It proposes a structured methodology for validating forensic tools and processes by u...