Phantom: General Trigger Attacks on Retrieval Augmented Language Generation

Retrieval Augmented Generation (RAG) expands the capabilities of modern large language models (LLMs) in chatbot applications, enabling developers to adapt and personalize the LLM output without expensive training or fine-tuning. RAG systems use an external knowledge database to retrieve the most relevant documents for a given query, providing this context to the LLM generator. While RAG achieves impressive utility in many applications, its adoption to enable personalized generative models introduces new security risks. In this work, we propose new attack surfaces for an adversary to compromise a victim's RAG system, by injecting a single malicious document in its knowledge database. We design Phantom, general two-step attack framework against RAG augmented LLMs. The first step involves crafting a poisoned document designed to be retrieved by the RAG system within the top-k results only when an adversarial trigger, a specific sequence of words acting as backdoor, is present in the victim's queries. In the second step, a specially crafted adversarial string within the poisoned document triggers various adversarial attacks in the LLM generator, including denial of service, reputation damage, privacy violations, and harmful behaviors. We demonstrate our attacks on multiple LLM architectures, including Gemma, Vicuna, and Llama.

SABRE: Robust Bayesian Peer-to-Peer Federated Learning

We introduce SABRE, a novel framework for robust variational Bayesian peer-to-peer federated learning. We analyze the robustness of the known variational Bayesian peer-to-peer federated learning framework (BayP2PFL) against poisoning attacks and subsequently show that BayP2PFL is not robust against those attacks. The new SABRE aggregation methodology is then devised to overcome the limitations of the existing frameworks. SABRE works well in non-IID settings, does not require the majority of the benign nodes over the compromised ones, and even outperforms the baseline algorithm in benign settings. We theoretically prove the robustness of our algorithm against data / model poisoning attacks in a decentralized linear regression setting. Proof-of-Concept evaluations on benchmark data from image classification demonstrate the superiority of SABRE over the existing frameworks under various poisoning attacks.

Experimental Security Analysis of DNN-based Adaptive Cruise Control under Context-Aware Perception Attacks

Adaptive Cruise Control (ACC) is a widely used driver assistance feature for maintaining desired speed and safe distance to the leading vehicles. This paper evaluates the security of the deep neural network (DNN) based ACC systems under stealthy perception attacks that strategically inject perturbations into camera data to cause forward collisions. We present a combined knowledge-and-data-driven approach to design a context-aware strategy for the selection of the most critical times for triggering the attacks and a novel optimization-based method for the adaptive generation of image perturbations at run-time. We evaluate the effectiveness of the proposed attack using an actual driving dataset and a realistic simulation platform with the control software from a production ACC system and a physical-world driving simulator while considering interventions by the driver and safety features such as Automatic Emergency Braking (AEB) and Forward Collision Warning (FCW). Experimental results show that the proposed attack achieves 142.9x higher success rate in causing accidents than random attacks and is mitigated 89.6% less by the safety features while being stealthy and robust to real-world factors and dynamic changes in the environment. This study provides insights into the role of human operators and basic safety interventions in preventing attacks.

Backdoor Attacks in Peer-to-Peer Federated Learning

We study backdoor attacks in peer-to-peer federated learning systems on different graph topologies and datasets. We show that only 5% attacker nodes are sufficient to perform a backdoor attack with 42% attack success without decreasing the accuracy on clean data by more than 2%. We also demonstrate that the attack can be amplified by the attacker crashing a small number of nodes. We evaluate defenses proposed in the context of centralized federated learning and show they are ineffective in peer-to-peer settings. Finally, we propose a defense that mitigates the attacks by applying different clipping norms to the model updates received from peers and local model trained by a node.

Byzantine Resilience at Swarm Scale: A Decentralized Blocklist Protocol from Inter-robot Accusations

The Weighted-Mean Subsequence Reduced (W-MSR) algorithm, the state-of-the-art method for Byzantine-resilient design of decentralized multi-robot systems, is based on discarding outliers received over Linear Consensus Protocol (LCP). Although W-MSR provides well-understood theoretical guarantees relating robust network connectivity to the convergence of the underlying consensus, the method comes with several limitations preventing its use at scale: (1) the number of Byzantine robots, F, to tolerate should be known a priori, (2) the requirement that each robot maintains 2F+1 neighbors is impractical for large F, (3) information propagation is hindered by the requirement that F+1 robots independently make local measurements of the consensus property in order for the swarm's decision to change, and (4) W-MSR is specific to LCP and does not generalize to applications not implemented over LCP. In this work, we propose a Decentralized Blocklist Protocol (DBP) based on inter-robot accusations. Accusations are made on the basis of locally-made observations of misbehavior, and once shared by cooperative robots across the network are used as input to a graph matching algorithm that computes a blocklist. DBP generalizes to applications not implemented via LCP, is adaptive to the number of Byzantine robots, and allows for fast information propagation through the multi-robot system while simultaneously reducing the required network connectivity relative to W-MSR. On LCP-type applications, DBP reduces the worst-case connectivity requirement of W-MSR from (2F+1)-connected to (F+1)-connected and the number of cooperative observers required to propagate new information from F+1 to just 1 observer. We demonstrate empirically that our approach to Byzantine resilience scales to hundreds of robots on cooperative target tracking, time synchronization, and localization case studies.

Network-Level Adversaries in Federated Learning

Federated learning is a popular strategy for training models on distributed, sensitive data, while preserving data privacy. Prior work identified a range of security threats on federated learning protocols that poison the data or the model. However, federated learning is a networked system where the communication between clients and server plays a critical role for the learning task performance. We highlight how communication introduces another vulnerability surface in federated learning and study the impact of network-level adversaries on training federated learning models. We show that attackers dropping the network traffic from carefully selected clients can significantly decrease model accuracy on a target population. Moreover, we show that a coordinated poisoning campaign from a few clients can amplify the dropping attacks. Finally, we develop a server-side defense which mitigates the impact of our attacks by identifying and up-sampling clients likely to positively contribute towards target accuracy. We comprehensively evaluate our attacks and defenses on three datasets, assuming encrypted communication channels and attackers with partial visibility of the network.

Automated Attack Synthesis by Extracting Finite State Machines from Protocol Specification Documents

Automated attack discovery techniques, such as attacker synthesis or model-based fuzzing, provide powerful ways to ensure network protocols operate correctly and securely. Such techniques, in general, require a formal representation of the protocol, often in the form of a finite state machine (FSM). Unfortunately, many protocols are only described in English prose, and implementing even a simple network protocol as an FSM is time-consuming and prone to subtle logical errors. Automatically extracting protocol FSMs from documentation can significantly contribute to increased use of these techniques and result in more robust and secure protocol implementations. In this work we focus on attacker synthesis as a representative technique for protocol security, and on RFCs as a representative format for protocol prose description. Unlike other works that rely on rule-based approaches or use off-the-shelf NLP tools directly, we suggest a data-driven approach for extracting FSMs from RFC documents. Specifically, we use a hybrid approach consisting of three key steps: (1) large-scale word-representation learning for technical language, (2) focused zero-shot learning for mapping protocol text to a protocol-independent information language, and (3) rule-based mapping from protocol-independent information to a specific protocol FSM. We show the generalizability of our FSM extraction by using the RFCs for six different protocols: BGPv4, DCCP, LTP, PPTP, SCTP and TCP. We demonstrate how automated extraction of an FSM from an RFC can be applied to the synthesis of attacks, with TCP and DCCP as case-studies. Our approach shows that it is possible to automate attacker synthesis against protocols by using textual specifications such as RFCs.

Are Self-Driving Cars Secure? Evasion Attacks against Deep Neural Networks for Steering Angle Prediction

Deep Neural Networks (DNNs) have tremendous potential in advancing the vision for self-driving cars. However, the security of DNN models in this context leads to major safety implications and needs to be better understood. We consider the case study of steering angle prediction from camera images, using the dataset from the 2014 Udacity challenge. We demonstrate for the first time adversarial testing-time attacks for this application for both classification and regression settings. We show that minor modifications to the camera image (an L2 distance of 0.82 for one of the considered models) result in mis-classification of an image to any class of attacker's choice. Furthermore, our regression attack results in a significant increase in Mean Square Error (MSE) by a factor of 69 in the worst case.

Leveraging Textual Specifications for Grammar-based Fuzzing of Network Protocols

Grammar-based fuzzing is a technique used to find software vulnerabilities by injecting well-formed inputs generated following rules that encode application semantics. Most grammar-based fuzzers for network protocols rely on human experts to manually specify these rules. In this work we study automated learning of protocol rules from textual specifications (i.e. RFCs). We evaluate the automatically extracted protocol rules by applying them to a state-of-the-art fuzzer for transport protocols and show that it leads to a smaller number of test cases while finding the same attacks as the system that uses manually specified rules.

On the Intriguing Connections of Regularization, Input Gradients and Transferability of Evasion and Poisoning Attacks

Transferability captures the ability of an attack against a machine-learning model to be effective against a different, potentially unknown, model. Studying transferability of attacks has gained interest in the last years due to the deployment of cyber-attack detection services based on machine learning. For these applications of machine learning, service providers avoid disclosing information about their machine-learning algorithms. As a result, attackers trying to bypass detection are forced to craft their attacks against a surrogate model instead of the actual target model used by the service. While previous work has shown that finding test-time transferable attack samples is possible, it is not well understood how an attacker may construct adversarial examples that are likely to transfer against different models, in particular in the case of training-time poisoning attacks. In this paper, we present the first empirical analysis aimed to investigate the transferability of both test-time evasion and training-time poisoning attacks. We provide a unifying, formal definition of transferability of such attacks and show how it relates to the input gradients of the surrogate and of the target classification models. We assess to which extent some of the most well-known machine-learning systems are vulnerable to transfer attacks, and explain why such attacks succeed (or not) across different models. To this end, we leverage some interesting connections highlighted in this work among the adversarial vulnerability of machine-learning models, their regularization hyperparameters and input gradients.