AIBearisharXiv – CS AI · Jun 97/10
🧠Researchers demonstrate a novel data poisoning attack targeting world models used in robot learning pipelines, showing how malicious prompts or dynamics hidden in training data can be activated only when processed through world models to generate unsafe robotic policies. The attack bypasses traditional safety measures by appearing benign in ground truth datasets while compromising downstream robot learning systems, affecting both action-conditioned and text-conditioned models.
AIBearisharXiv – CS AI · Jun 57/10
🧠Researchers propose the first formal threat model for Retrieval-Augmented Generation (RAG) systems, which combine LLMs with external document retrieval. The framework identifies new security vulnerabilities including document membership inference and data poisoning attacks that emerge from RAG's reliance on external knowledge bases, addressing a critical gap in AI safety research.
AIBearisharXiv – CS AI · May 277/10
🧠Researchers have identified a new data poisoning vulnerability in large language models called 'covert control attacks' that uses semantic associations to hide malicious instructions rather than obvious trigger phrases. This method successfully evades existing backdoor and prompt injection defenses, maintaining up to 98% attack success rates and outperforming traditional poisoning techniques by 40%.
AIBearisharXiv – CS AI · Apr 107/10
🧠Researchers have demonstrated the first multi-targeted backdoor attack against graph neural networks (GNNs) in graph classification tasks, using a novel subgraph injection method that simultaneously redirects multiple predictions to different target labels while maintaining clean accuracy. The attack shows high efficacy across multiple GNN architectures and datasets, with resilience against existing defense mechanisms, exposing significant vulnerabilities in GNN security.
AINeutralarXiv – CS AI · Mar 57/10
🧠Researchers propose a new method called Mutual Information Unlearnable Examples (MI-UE) to protect data privacy by preventing unauthorized AI models from learning from scraped data. The approach uses mutual information theory to create more effective data poisoning techniques that impede deep learning model generalization.
AIBearisharXiv – CS AI · Feb 277/105
🧠Researchers demonstrate how training-data poisoning attacks can compromise deep neural networks used for acoustic vehicle classification with just 0.5% corrupted data, achieving 95.7% attack success rate while remaining undetectable. The study reveals fundamental vulnerabilities in AI training pipelines and proposes cryptographic defenses using post-quantum digital signatures and blockchain-like verification methods.
AINeutralarXiv – CS AI · Jun 96/10
🧠Researchers propose Safe-RULE, a new reinforcement unlearning framework designed to defend offline safe reinforcement learning systems against data poisoning attacks. The approach removes malicious data influence without requiring model retraining or access to original training environments, addressing a critical vulnerability in safety-critical applications like robotics.
