Search

Multimodal Large Language Models (MLLMs) demonstrate remarkable capabilities that increasingly influence various aspects of our daily lives, constantly defining the new boundary of Artificial General Intelligence (AGI). Image modalities, enriched with profound semantic information and a more continuous mathematical nature compared to other modalities, greatly enhance the functionalities of MLLMs when integrated. However, this integration serves as a double-edged sword, providing attackers with expansive vulnerabilities to exploit for highly covert and harmful attacks. The pursuit of reliable AI systems like powerful MLLMs has emerged as a pivotal area of contemporary research. In this paper, we endeavor to demostrate the multifaceted risks associated with the incorporation of image modalities into MLLMs. Initially, we delineate the foundational components and training processes of MLLMs. Subsequently, we construct a threat model, outlining the security vulnerabilities intrinsic to MLLMs. Moreover, we analyze and summarize existing scholarly discourses on MLLMs' attack and defense mechanisms, culminating in suggestions for the future research on MLLM security. Through this comprehensive analysis, we aim to deepen the academic understanding of MLLM security challenges and propel forward the development of trustworthy MLLM systems., Comment: 8 pages, 1 figure

Existing works have extensively studied adversarial examples, which are minimal perturbations that can mislead the output of deep neural networks (DNNs) while remaining imperceptible to humans. However, in this work, we reveal the existence of a harmless perturbation space, in which perturbations drawn from this space, regardless of their magnitudes, leave the network output unchanged when applied to inputs. Essentially, the harmless perturbation space emerges from the usage of non-injective functions (linear or non-linear layers) within DNNs, enabling multiple distinct inputs to be mapped to the same output. For linear layers with input dimensions exceeding output dimensions, any linear combination of the orthogonal bases of the nullspace of the parameter consistently yields no change in their output. For non-linear layers, the harmless perturbation space may expand, depending on the properties of the layers and input samples. Inspired by this property of DNNs, we solve for a family of general perturbation spaces that are redundant for the DNN's decision, and can be used to hide sensitive data and serve as a means of model identification. Our work highlights the distinctive robustness of DNNs (i.e., consistency under large magnitude perturbations) in contrast to adversarial examples (vulnerability for small imperceptible noises).

Open-source Large Language Models (LLMs) have recently gained popularity because of their comparable performance to proprietary LLMs. To efficiently fulfill domain-specialized tasks, open-source LLMs can be refined, without expensive accelerators, using low-rank adapters. However, it is still unknown whether low-rank adapters can be exploited to control LLMs. To address this gap, we demonstrate that an infected adapter can induce, on specific triggers, an LLM to output content defined by an adversary and to even maliciously use tools. To train a Trojan adapter, we propose two novel attacks, POLISHED and FUSION, that improve over prior approaches. POLISHED uses LLM-enhanced paraphrasing to polish benchmark poisoned datasets. In contrast, in the absence of a dataset, FUSION leverages an over-poisoning procedure to transform a benign adaptor. In our experiments, we first conduct two case studies to demonstrate that a compromised LLM agent can execute malware to control system (e.g., LLM-driven robot) or launch a spear-phishing attack. Then, in terms of targeted misinformation, we show that our attacks provide higher attack effectiveness than the baseline and, for the purpose of attracting downloads, preserve or improve the adapter's utility. Finally, we design and evaluate three potential defenses, yet none proved entirely effective in safeguarding against our attacks.

We propose a digital twin-assisted adaptive preloading scheme to enhance bandwidth efficiency and user quality of experience (QoE) in short video streaming. We first analyze the relationship between the achievable throughput and video bitrate and critical factors that affect the preloading decision, including the buffer size and bitrate selection. We then construct a digital twin-assisted adaptive preloading framework for short video streaming. By collecting and analyzing historical throughput and tracking behavior information, a throughput prediction model and a probabilistic model can be constructed to accurately predict future throughput and user behavior, respectively. Using the predicted information and real-time running status data from a short video application, we design a preloading strategy to enhance bandwidth efficiency while guaranteeing user QoE. Simulation results demonstrate the effectiveness of our proposed scheme comparing with the state-of-the-art preloading schemes.

In this paper, we propose a novel distributed learning scheme, named group-based split federated learning (GSFL), to speed up artificial intelligence (AI) model training. Specifically, the GSFL operates in a split-then-federated manner, which consists of three steps: 1) Model distribution, in which the access point (AP) splits the AI models and distributes the client-side models to clients; 2) Model training, in which each client executes forward propagation and transmit the smashed data to the edge server. The edge server executes forward and backward propagation and then returns the gradient to the clients for updating local client-side models; and 3) Model aggregation, in which edge servers aggregate the server-side and client-side models. Simulation results show that the GSFL outperforms vanilla split learning and federated learning schemes in terms of overall training latency while achieving satisfactory accuracy.

In this paper, we propose a novel and practical mechanism which enables the service provider to verify whether a suspect model is stolen from the victim model via model extraction attacks. Our key insight is that the profile of a DNN model's decision boundary can be uniquely characterized by its Universal Adversarial Perturbations (UAPs). UAPs belong to a low-dimensional subspace and piracy models' subspaces are more consistent with victim model's subspace compared with non-piracy model. Based on this, we propose a UAP fingerprinting method for DNN models and train an encoder via contrastive learning that takes fingerprint as inputs, outputs a similarity score. Extensive studies show that our framework can detect model IP breaches with confidence > 99.99 within only 20 fingerprints of the suspect model. It has good generalizability across different model architectures and is robust against post-modifications on stolen models., Comment: Accepted to CVPR 2022 (Oral Presentation)

Numerous open-source and commercial malware detectors are available. However, their efficacy is threatened by new adversarial attacks, whereby malware attempts to evade detection, e.g., by performing feature-space manipulation. In this work, we propose an explainability-guided and model-agnostic framework for measuring the efficacy of malware detectors when confronted with adversarial attacks. The framework introduces the concept of Accrued Malicious Magnitude (AMM) to identify which malware features should be manipulated to maximize the likelihood of evading detection. We then use this framework to test several state-of-the-art malware detectors' ability to detect manipulated malware. We find that (i) commercial antivirus engines are vulnerable to AMM-guided manipulated samples; (ii) the ability of a manipulated malware generated using one detector to evade detection by another detector (i.e., transferability) depends on the overlap of features with large AMM values between the different detectors; and (iii) AMM values effectively measure the importance of features and explain the ability to evade detection. Our findings shed light on the weaknesses of current malware detectors, as well as how they can be improved.

Natural language processing (NLP) systems have been proven to be vulnerable to backdoor attacks, whereby hidden features (backdoors) are trained into a language model and may only be activated by specific inputs (called triggers), to trick the model into producing unexpected behaviors. In this paper, we create covert and natural triggers for textual backdoor attacks, \textit{hidden backdoors}, where triggers can fool both modern language models and human inspection. We deploy our hidden backdoors through two state-of-the-art trigger embedding methods. The first approach via homograph replacement, embeds the trigger into deep neural networks through the visual spoofing of lookalike character replacement. The second approach uses subtle differences between text generated by language models and real natural text to produce trigger sentences with correct grammar and high fluency. We demonstrate that the proposed hidden backdoors can be effective across three downstream security-critical NLP tasks, representative of modern human-centric NLP systems, including toxic comment detection, neural machine translation (NMT), and question answering (QA). Our two hidden backdoor attacks can achieve an Attack Success Rate (ASR) of at least $97\%$ with an injection rate of only $3\%$ in toxic comment detection, $95.1\%$ ASR in NMT with less than $0.5\%$ injected data, and finally $91.12\%$ ASR against QA updated with only 27 poisoning data samples on a model previously trained with 92,024 samples (0.029\%). We are able to demonstrate the adversary's high success rate of attacks, while maintaining functionality for regular users, with triggers inconspicuous by the human administrators.

Intuitively, a backdoor attack against Deep Neural Networks (DNNs) is to inject hidden malicious behaviors into DNNs such that the backdoor model behaves legitimately for benign inputs, yet invokes a predefined malicious behavior when its input contains a malicious trigger. The trigger can take a plethora of forms, including a special object present in the image (e.g., a yellow pad), a shape filled with custom textures (e.g., logos with particular colors) or even image-wide stylizations with special filters (e.g., images altered by Nashville or Gotham filters). These filters can be applied to the original image by replacing or perturbing a set of image pixels.

Deep neural networks (DNNs) have been proven vulnerable to backdoor attacks, where hidden features (patterns) trained to a normal model, which is only activated by some specific input (called triggers), trick the model into producing unexpected behavior. In this paper, we create covert and scattered triggers for backdoor attacks, invisible backdoors, where triggers can fool both DNN models and human inspection. We apply our invisible backdoors through two state-of-the-art methods of embedding triggers for backdoor attacks. The first approach on Badnets embeds the trigger into DNNs through steganography. The second approach of a trojan attack uses two types of additional regularization terms to generate the triggers with irregular shape and size. We use the Attack Success Rate and Functionality to measure the performance of our attacks. We introduce two novel definitions of invisibility for human perception; one is conceptualized by the Perceptual Adversarial Similarity Score (PASS) and the other is Learned Perceptual Image Patch Similarity (LPIPS). We show that the proposed invisible backdoors can be fairly effective across various DNN models as well as four datasets MNIST, CIFAR-10, CIFAR-100, and GTSRB, by measuring their attack success rates for the adversary, functionality for the normal users, and invisibility scores for the administrators. We finally argue that the proposed invisible backdoor attacks can effectively thwart the state-of-the-art trojan backdoor detection approaches, such as Neural Cleanse and TABOR.