Cybersecurity & Exploits

Global: Researchers Reveal Universal Encoder-Level Attack on Audio Language ModelsOn December 29, 2025, a team of researchers—Roee Ziv, Raz Lapid, and Moshe Sipper—published a study describing a universal targeted latent‑space…

Global: Researchers Reveal Efficient Backdoor Technique Targeting Diffusion Model BlocksA new attack method named BadBlocks enables adversaries to embed covert triggers into diffusion‑based image generators by compromising selected blocks of…

Global: SyncGait Enables Long-Distance Drone Authentication via Implicit GaitResearchers led by Zijian Ling announced a new mutual authentication system for drone delivery on December 29, 2025, that leverages a user’s…

Global: New Benchmark Reveals Prompt-Based Over-Generation Attacks on LLMsResearchers introduced a benchmark that evaluates black‑box, query‑only prompt attacks capable of forcing large language models (LLMs) to generate far beyond their…

Global: Adversarial Prompts Exploit Load Imbalance in Mixture-of-Experts Language ModelsResearchers have identified a denial-of-service vulnerability in large language models that employ Mixture-of-Experts (MoE) architectures. The study, posted to arXiv in…

Global: Meta-Learning Framework Achieves Near-Perfect Malware Detection AccuracyA new study released on arXiv introduces Meta Learning Malware Detection (MeLeMaD), a framework that applies Model-Agnostic Meta-Learning (MAML) to improve the detection…

Global: Visual-Based Spam Filter Achieves Over 98% Detection AccuracyA new visual-based spam detection system has been introduced by researchers, aiming to counter sophisticated email obfuscation methods. The study, posted to…

Global: Study Finds Large Language Models Ineffective for Password CrackingIn October 2025, researchers posted a paper on arXiv that examined whether state‑of‑the‑art open‑source large language models could be leveraged to…

Global: Unleash Protocol Hack Leads to $3.9 Million TheftSecurity Breach OverviewA security breach resulted in the theft of approximately $3.9 million from Unleash Protocol, a platform that enables creators to…

Global: Equacode Demonstrates High Success Rates in Multi‑Strategy LLM JailbreaksResearchers have introduced a new multi‑strategy jailbreak method for large language models, called Equacode, which converts malicious prompts into mathematical equations…