Academic papers, new techniques, benchmarks, and theoretical findings in AI/LLM security.
This research addresses security challenges in Internet of Things (IoT) devices by improving radio frequency fingerprint identification (RFFI, a method that uniquely identifies devices based on their wireless signal characteristics) using federated learning (a distributed AI training approach where data stays on local devices rather than being sent to a central server). The paper proposes a feature alignment strategy to handle non-IID data (data that isn't uniformly distributed across different receivers), which occurs when different receivers have different hardware and environmental conditions, and demonstrates that the approach achieves 90.83% identification accuracy with improved stability compared to existing federated learning methods.
Fix: The paper proposes a feature alignment strategy based on federated learning that guides each client (receiver) to learn aligned intermediate feature representations during local training, effectively mitigating the adverse impact of distribution shifts on model generalization in heterogeneous wireless environments.
IEEE Xplore (Security & AI Journals)QuEST is a new framework that makes backdoor attacks (hidden malicious behaviors injected into AI models) more stealthy and efficient when models undergo quantization (compressing models to use less memory and computation). The framework uses special training techniques and parameter sharing to hide the attack from detection systems while reducing the computational resources needed to carry out the attack.
Large vision-language models (LVLMs, which are AIs that understand both images and text) can be attacked using simple visual transformations, such as rotations or color changes, that fool them into giving wrong answers. Researchers found that combining multiple harmful transformations can make these attacks more effective, and they can be optimized using gradient approximation (a mathematical technique to find the best attack parameters). This research highlights a previously overlooked safety risk in how well LVLMs resist these kinds of adversarial attacks (attempts to trick AI systems).
Researchers discovered a new attack called Lure that targets generative language models (GLMs, which are AI systems that generate text) during the fine-tuning process (when developers customize an open-source model with their own data). By hiding malicious code in the source code of an open-source model, attackers can trick a fine-tuned model into remembering and later revealing the proprietary data used to customize it through specially crafted prompts (input text designed to trigger specific outputs).
Large Language Models (LLMs, AI systems trained on massive amounts of text) used in task-oriented dialogue systems (AI assistants designed to help users complete specific goals like booking travel) can accidentally memorize and leak sensitive training data, including personal information like phone numbers and complete travel schedules. Researchers demonstrated new attack techniques that can extract thousands of pieces of training data from these systems with over 70% accuracy in the best cases. The paper identifies factors that influence how much data LLMs memorize in dialogue systems but does not propose specific fixes.
This research paper addresses security risks from synthetic tabular data (AI-generated fake datasets) by proposing PBC-TabFip, a fingerprinting framework that embeds hidden identifiers into synthetic data to detect unauthorized copying and identify who leaked it. The framework uses diffusion models (AI systems that generate data by gradually refining random noise) and Tardos codes (a mathematical scheme for tracking which user leaked protected content) to protect synthetic tables even when primary keys (unique identifiers for database rows) are missing or altered, and to resist collusion attacks (when multiple users combine their copies to remove the fingerprint).
Fix: The source proposes 'PBC-TabFip' as the solution: a framework that 'readily incorporates with symmetric Tardos codes of arbitrary alphabet sizes' to enable fingerprinting of synthetic tabular data generated by diffusion models. The paper also proposes specific schemes including 'binary TabFip and TabFip+, quaternary TabFip* and TabFip+*' that use 'Bit Matching (BM) and Valid Bit Matching (VBM) mechanisms' to identify malicious users. According to the authors, 'TabFip with Tardos codes identifies at least one of the colluders with 100% probability and without detecting innocent against two types of collusion attack.'
IEEE Xplore (Security & AI Journals)Individual large language models (LLMs, AI systems trained to generate text or code) often produce inconsistent outputs and show biases, which limits their usefulness. This survey examines ensemble learning techniques (methods that combine multiple models to improve results), categorizing seven approaches like weight merging (combining model parameters), mixture-of-experts (routing inputs to specialized models), and output ensemble (combining multiple model outputs), to show how combining LLMs can improve output quality and diversity in both text and code generation.
This research proposes a framework to improve physical layer security (PLS, the protection of wireless communications at the hardware level) in UAV-aided backscatter networks, which are systems where drones help passive devices transmit data efficiently. The framework uses federated learning (FL, a technique where multiple agents learn from data without sharing raw information) combined with deep reinforcement learning (DRL, machine learning where agents learn by trial and error to make optimal decisions) to optimize how a UAV positions itself, allocates power, and manages a backscatter tag while protecting against eavesdroppers by injecting artificial noise.
Diffusion-based image editing systems (AI tools that modify images based on text descriptions) can be manipulated maliciously, and while adding imperceptible perturbations (tiny, invisible changes) to images helps protect against this, existing defenses don't work well across different models. This paper proposes TDAE, a system that combines image and text-based defenses to create images that are harder to maliciously edit, even when attacked by unfamiliar editing models.
This research proposes a new authentication and key agreement (AKA, a process where devices verify each other's identity and create shared secret keys for secure communication) scheme for VANETs (vehicular ad hoc networks, where cars communicate directly with each other without central infrastructure). The scheme uses a consortium blockchain (a shared, distributed ledger controlled by a group of organizations rather than one central authority) to work in asynchronous environments, where messages may arrive out of order or with delays, and employs lightweight cryptographic techniques (mathematical methods that require less computing power) to reduce system overhead.
Adversarial SQL injection (SQLi, a technique where attackers modify their attacks based on feedback from a Web Application Firewall to bypass it) has become a serious threat, with automated tools like AdvSQLi and GPTFuzzer making it easier to find vulnerabilities. The paper proposes a hybrid defense system combining Character-Level CNN (a neural network that analyzes attack payloads character-by-character to find harmful patterns) and Reinforcement Learning (a type of AI training that learns through trial and feedback) to detect these advanced attacks, showing that this approach can catch malicious patterns even when attackers try to disguise their payloads.
This research proposes ATRIA, a system for verifying that copies of data stored across multiple edge computing servers are authentic and haven't been tampered with. ATRIA uses TEEs (trusted execution environments, which are secure hardware areas that isolate sensitive operations) to shift the work of generating verification tags from resource-limited user devices to more powerful servers, while also protecting user privacy through anonymous identities that a trusted authority can trace if needed. The system protects against attacks where servers collude or create fake data on demand, and testing shows it uses less computing power than similar existing approaches.
Bitcoin is shifting from system rewards to transaction fees (payments users include with their transactions) to incentivize miners, but this creates a 'mining gap' where miners turn off their equipment when fees are too low, weakening Bitcoin's security. This paper identifies this as an 'egoistic dilemma' where both users and miners act selfishly, and proposes an incentive mechanism based on zero-determinant theory (a game theory approach) to solve the problem.
Wireless Sensor Networks (WSNs, collections of small wireless devices that sense and relay data) are vulnerable to node failures and malicious attacks because they operate with limited resources in open environments. This paper proposes EFTE, a framework that evaluates the trustworthiness of individual nodes by measuring their communication quality, remaining battery power, behavior consistency, and movement patterns, then uses entropy-based weighting (a mathematical approach to handle uncertainty in data) and a fuzzy inference system (a method that makes decisions from incomplete or uncertain information) to identify and isolate untrustworthy nodes while protecting data with lightweight encryption.
This paper introduces CipheRAG, a system that helps large language models (LLMs) safely use external knowledge sources while protecting sensitive data. The system balances two competing needs: keeping data private while still retrieving information quickly, which existing approaches struggle to do because cryptography-based methods are slow while faster methods leak more information.
Camouflaged object detection (COD, identifying objects that blend into their surroundings due to similar colors and textures) is difficult because current methods struggle to precisely outline object boundaries. Researchers propose BADANet, a neural network architecture that combines boundary detection with distraction mining (techniques to ignore confusing background features) using specialized modules like a boundary-aware distracted attention block to improve detection accuracy. The method was tested on four datasets and outperformed 18 existing approaches.
This research addresses how to safely explore environments using reinforcement learning (RL, a type of AI training where a system learns by trial and error) without causing damage or violating safety rules. The paper introduces safe equilibrium exploration (SEE), a method that balances two competing goals: expanding the area where exploration is allowed (the feasible zone) and building a more accurate model of how the environment works, showing that these two objectives improve each other and can reach an optimal balance without any safety violations.
Contrastive learning (a machine learning technique where the AI learns to group similar items together and push different items apart) can suffer from sampling bias when similar samples belong to different classes or dissimilar samples belong to the same class, hurting classification accuracy. This paper proposes using out-of-distribution (OOD) detection, which identifies and masks unusual or misclassified samples, to create a better contrastive learning model that can work without needing a separate collection of known unusual samples. The authors generate synthetic samples at the boundary between normal and unusual data to train an improved detector that produces more reliable classifications.
Researchers proposed Adaptive Token Dictionary (ATD), a new transformer architecture (a type of AI model good at learning relationships between different parts of data) designed to improve image restoration tasks like super-resolution and denoising while reducing computational demands. Unlike traditional transformers that struggle with high computational costs, ATD uses a learnable token dictionary (a set of learned patterns representing typical image structures) and a cross-attention mechanism (a way for the model to compare input data against these learned patterns) to achieve better performance with lower computational complexity.
AIRPNet is a new AI system that restores damaged images while keeping them hidden from cloud services, protecting user privacy. The system works by concealing low-quality images inside other images using a technique called steganography (hiding data within other data), then restoring the hidden image without ever exposing it during processing. This approach offers better privacy protection than existing methods while maintaining image quality.