Recent advancements in Reinforcement Learning with Human Feedback (RLHF) have significantly impacted the alignment of Large Language Models (LLMs). The sensitivity of reinforcement learning algorithms such as Proximal Policy Optimization (PPO) has led to new line work on Direct Policy Optimization (DPO), which treats RLHF in a supervised learning framework. The increased practical use of these RLHF methods warrants an analysis of their vulnerabilities. In this work, we investigate the vulnerabilities of DPO to poisoning attacks under different scenarios and compare the effectiveness of preference poisoning, a first of its kind. We comprehensively analyze DPO's vulnerabilities under different types of attacks, i.e., backdoor and non-backdoor attacks, and different poisoning methods across a wide array of language models, i.e., LLama 7B, Mistral 7B, and Gemma 7B. We find that unlike PPO-based methods, which, when it comes to backdoor attacks, require at least 4\% of the data to be poisoned to elicit harmful behavior, we exploit the true vulnerabilities of DPO more simply so we can poison the model with only as much as 0.5\% of the data. We further investigate the potential reasons behind the vulnerability and how well this vulnerability translates into backdoor vs non-backdoor attacks.

Federated learning (FL) has become a popular tool for solving traditional Reinforcement Learning (RL) tasks. The multi-agent structure addresses the major concern of data-hungry in traditional RL, while the federated mechanism protects the data privacy of individual agents. However, the federated mechanism also exposes the system to poisoning by malicious agents that can mislead the trained policy. Despite the advantage brought by FL, the vulnerability of Federated Reinforcement Learning (FRL) has not been well-studied before. In this work, we propose a general framework to characterize FRL poisoning as an optimization problem and design a poisoning protocol that can be applied to policy-based FRL. Our framework can also be extended to FRL with actor-critic as a local RL algorithm by training a pair of private and public critics. We provably show that our method can strictly hurt the global objective. We verify our poisoning effectiveness by conducting extensive experiments targeting mainstream RL algorithms and over various RL OpenAI Gym environments covering a wide range of difficulty levels. Within these experiments, we compare clean and baseline poisoning methods against our proposed framework. The results show that the proposed framework is successful in poisoning FRL systems and reducing performance across various environments and does so more effectively than baseline methods. Our work provides new insights into the vulnerability of FL in RL training and poses new challenges for designing robust FRL algorithms

Github Copilot, trained on billions of lines of public code, has recently become the buzzword in the computer science research and practice community. Although it is designed to provide powerful intelligence to help developers implement safe and effective code, practitioners and researchers raise concerns about its ethical and security problems, e.g., should the copyleft licensed code be freely leveraged or insecure code be considered for training in the first place? These problems pose a significant impact on Copilot and other similar products that aim to learn knowledge from large-scale source code through deep learning models, which are inevitably on the rise with the fast development of artificial intelligence. To mitigate such impacts, we argue that there is a need to invent effective mechanisms for protecting open-source code from being exploited by deep learning models. To this end, we design and implement a prototype, CoProtector, which utilizes data poisoning techniques to arm source code repositories for defending against such exploits. Our large-scale experiments empirically show that CoProtector is effective in achieving its purpose, significantly reducing the performance of Copilot-like deep learning models while being able to stably reveal the secretly embedded watermark backdoors.

In big data era, machine learning is one of fundamental techniques in intrusion detection systems (IDSs). However, practical IDSs generally update their decision module by feeding new data then retraining learning models in a periodical way. Hence, some attacks that comprise the data for training or testing classifiers significantly challenge the detecting capability of machine learning-based IDSs. Poisoning attack, which is one of the most recognized security threats towards machine learning-based IDSs, injects some adversarial samples into the training phase, inducing data drifting of training data and a significant performance decrease of target IDSs over testing data. In this paper, we adopt the Edge Pattern Detection (EPD) algorithm to design a novel poisoning method that attack against several machine learning algorithms used in IDSs. Specifically, we propose a boundary pattern detection algorithm to efficiently generate the points that are near to abnormal data but considered to be normal ones by current classifiers. Then, we introduce a Batch-EPD Boundary Pattern (BEBP) detection algorithm to overcome the limitation of the number of edge pattern points generated by EPD and to obtain more useful adversarial samples. Based on BEBP, we further present a moderate but effective poisoning method called chronic poisoning attack. Extensive experiments on synthetic and three real network data sets demonstrate the performance of the proposed poisoning method against several well-known machine learning algorithms and a practical intrusion detection method named FMIFS-LSSVM-IDS.