Research
Academic papers, new techniques, benchmarks, and theoretical findings in AI/LLM security.
Academic papers, new techniques, benchmarks, and theoretical findings in AI/LLM security.
2,033 items
Researchers present 3DGAA, a framework for physical adversarial attacks on camera-based perception in autonomous vehicles. It uses 3D multi-view optimization with a 3D Gaussian splatting surrogate to produce print-only vehicle wraps that keep geometry unchanged. In CARLA simulations and miniature-vehicle tests, the wraps substantially reduced detection confidence and average precision across multiple modern detectors.
Researchers present a CSI-based Wi-Fi sensing system that detects human presence using opportunistic ambient links from devices such as smartphones, laptops and IoT terminals. The method combines a reference-based subspace denoising step, which models long-term background statistics during unoccupied periods, with multi-region joint sensing that fuses room device-to-AP links and corridor AP-to-AP links. Evaluations in single-room and four-room university deployments report that denoising improves room sensing and multi-region fusion further improves presence inference.
PB-Net v2 is a revised dynamic hand gesture authentication model that refines data tailoring, the physiological and behavioral branches, and the feature fusion module. On SCUT-DHGA, SCUT-DHGA-br, and SCUT-RealDHGA, it achieves the lowest equal error rates among 22 state-of-the-art models under four evaluation protocols. It also shows improved robustness against imitation-oriented spoofing attacks under three such protocols.
The paper studies Federated Generalized Category Discovery (Fed-GCD), where clients collaborate privately to cluster unlabeled samples from known and unknown classes. It proposes Personalized Contrastive Graph Learning (PCGL), which separates generic and personalized knowledge, using a KNN-Former for local-graph contrastive learning and adaptive parameter masking with personalized aggregation. On six datasets, PCGL outperforms state-of-the-art methods, with a 4.8% average gain on All ACC for long-tailed natural and medical image datasets.
MITRE ATLAS v2026.08 is a release of the knowledge base of adversary tactics and techniques involving AI. It contains 1 matrix, 16 tactics, 114 techniques, 83 sub-techniques, 39 mitigations and 72 case studies. The update adds new techniques for autonomous attack orchestration, AI agent communication and AI agent tools, and renames the AI Attack Staging tactic to AI Attack Adaptation.
Researchers examined how boards and executives make cyber risk decisions, drawing on more than 1,800 simulation exercises and executive interviews. They found that decision quality falls sharply as threat complexity rises. The study identifies four managerial heuristics for strengthening cyber resilience: faster adaptation, iterative learning, business-wide responsibility for cyber risk, and proactively building capabilities.
The source is a bibliographic record for a paper titled "FlexDPI: Verifiable and privacy-preserving deep packet inspection with flexible rule subscription." It was published in September 2026 in the Journal of Information Security and Applications, Volume 101, by Xiaopin Lv, Kai Zhang, Jinguo Li, Lifei Wei and Jianting Ning. The source text contains no abstract, method or findings.
AuditML is a publicly auditable privacy-preserving framework for machine learning inference in the semi-honest model. It combines arithmetic secret sharing with commitment-based audit data, implemented on MP-SPDZ, and evaluated on linear regression, fixed-threshold binary logistic classification, and linear-kernel SVM. At a batch size of 50, linear and logistic inference completed online in under one second, while SVM took 10.0939–40.0334 seconds.
WPEBA is a frequency-driven ensemble black-box adversarial attack for visual recognition systems that uses wavelet packet decomposition to split images into frequency sub-bands. It adjusts sub-band weights from internal gradient feedback and updates surrogate-model weights from target-model query feedback, reaching an average attack success rate of nearly 99% with one or two queries across six standard architectures. The authors report it also remains effective against defended models and the Google Cloud Vision API.
GECOMP is a generative textual adversarial attack that uses reinforcement learning for policy optimization. An LLM-based generator rewrites input sequences using an extensible library of compositional perturbations, constrained by semantic similarity and edit magnitude. Across four public datasets and five victim models, it achieved higher attack success rates than ten baseline methods while using fewer queries and less edit magnitude.
ADA-MIA is a black-box membership inference attack against contrastive learning (CL) encoders. It needs only an auxiliary dataset of training samples, without their membership labels, and it constructs membership features from aggressively augmented views of the target sample, enabling unsupervised inference. Across four datasets, it matches the performance of seven existing attacks while requiring only black-box embedding access to the encoder.
A systematic review of security and privacy-preserving mechanisms in collaborative machine learning proposes a novel taxonomy for organizing them. The paper appeared in the Journal of Information Security and Applications, Volume 102, in November 2026. The authors are Ammar Yaseen Tuama, Taha Mohammed Hasan, and Ahmad Chalak Shakir.
Researchers at Bocconi University, with OpenAI Economic Research, ran an experiment with more than 1,000 first-year undergraduates on a real-world marketing case for the university's merchandise store. Students given ChatGPT (GPT-4o) access scored almost a full point higher on a five-point rubric, with more ideas, clearer logic and answers closer to expert recommendations. A causal-reasoning exercise, unrelated to AI, led students to produce a wider range of more distinct ideas, and students who received both interventions showed both effects.
This ACM Computing Surveys article, published in Volume 58, Issue 15 (pages 1-38) in November 2026, surveys security risks across AI systems, from large language models to AI agents and embodied agents. The source text provided contains only the bibliographic citation and no article content, so no findings or methods can be reported from it.