Fengqing Jiang, Yichen Feng, Yuetai Li, Luyao Niu, Basel Alomair, Radha Poovendran. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Yuetai Li, Zhangchen Xu, Fengqing Jiang, Bhaskar Ramasubramanian, Luyao Niu, Bill Yuchen Lin, Xiang Yue, Radha Poovendran. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large language models (LLMs) are expected to follow instructions from users and engage in conversations. Techniques to enhance LLMs' instruction-following capabilities typically fine-tune them using data structured according to a predefined chat template. Although chat templates are shown to be effective in optimizing LLM performance, their impact on safety alignment of LLMs has been less understood, which is crucial for deploying LLMs safely at scale. In this paper, we investigate how chat templates affect safety alignment of LLMs. We identify a common vulnerability, named ChatBug, that is introduced by chat templates. Our key insight to identify ChatBug is that the chat templates provide a rigid format that need to be followed by LLMs, but not by users. Hence, a malicious user may not necessarily follow the chat template when prompting LLMs. Instead, malicious users could leverage their knowledge of the chat template and accordingly craft their prompts to bypass safety alignments of LLMs. We study two attacks to exploit the ChatBug vulnerability. Additionally, we demonstrate that the success of multiple existing attacks can be attributed to the ChatBug vulnerability. We show that a malicious user can exploit the ChatBug vulnerability of eight state-of-the-art (SOTA) LLMs and effectively elicit unintended responses from these models. Moreover, we show that ChatBug can be exploited by existing jailbreak attacks to enhance their attack success rates. We investigate potential countermeasures to ChatBug. Our results show that while adversarial training effectively mitigates the ChatBug vulnerability, the victim model incurs significant performance degradation. These results highlight the trade-off between safety alignment and helpfulness. Developing new methods for instruction tuning to balance this trade-off is an open and critical direction for future research.

High-quality instruction data is critical for aligning large language models (LLMs). Although some models, such as Llama-3-Instruct, have open weights, their alignment data remain private, which hinders the democratization of AI. High human labor costs and a limited, predefined scope for prompting prevent existing open-source data creation methods from scaling effectively, potentially limiting the diversity and quality of public alignment datasets. Is it possible to synthesize high-quality instruction data at scale by extracting it directly from an aligned LLM? We present a self-synthesis method for generating large-scale alignment data named Magpie. Our key observation is that aligned LLMs like Llama-3-Instruct can generate a user query when we input only the pre-query templates up to the position reserved for user messages, thanks to their auto-regressive nature. We use this method to prompt Llama-3-Instruct and generate 4 million instructions along with their corresponding responses. We further introduce extensions of Magpie for filtering, generating multi-turn, preference optimization, domain-specific and multilingual datasets. We perform a comprehensive analysis of the Magpie-generated data. To compare Magpie-generated data with other public instruction datasets (e.g., ShareGPT, WildChat, Evol-Instruct, UltraChat, OpenHermes, Tulu-V2-Mix, GenQA), we fine-tune Llama-3-8B-Base with each dataset and evaluate the performance of the fine-tuned models. Our results indicate that using Magpie for supervised fine-tuning (SFT) solely can surpass the performance of previous public datasets utilized for both SFT and preference optimization, such as direct preference optimization with UltraFeedback. We also show that in some tasks, models supervised fine-tuned with Magpie perform comparably to the official Llama-3-8B-Instruct, despite the latter being enhanced with 10 million data points through SFT and subsequent preference optimization. This advantage is evident on alignment benchmarks such as AlpacaEval, ArenaHard, and WildBench.

Zhangchen Xu, Fengqing Jiang, Luyao Niu, Bill Yuchen Lin, Radha Poovendran. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Fengqing Jiang, Zhangchen Xu, Luyao Niu, Zhen Xiang, Bhaskar Ramasubramanian, Bo Li, Radha Poovendran. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Zhangchen Xu, Fengqing Jiang, Luyao Niu, Jinyuan Jia, Bill Yuchen Lin, Radha Poovendran. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Compared with the traditional usage of large language models (LLMs) where users directly send queries to an LLM, LLM-integrated applications serve as middleware to refine users' queries with domain-specific knowledge to better inform LLMs and enhance the responses. However, LLM-integrated applications also introduce new attack surfaces. This work considers a setup where the user and LLM interact via an application in the middle. We focus on the interactions that begin with user's queries and end with LLM-integrated application returning responses to the queries, powered by LLMs at the service backend. We identify potential high-risk vulnerabilities in this setting that can originate from the malicious application developer or from an outsider threat initiator that can control the database access, manipulate and poison high-risk data for the user. Successful exploits of the identified vulnerabilities result in the users receiving responses tailored to the intent of a threat initiator. We assess such threats against LLM-integrated applications empowered by GPT-3.5 and GPT-4. Our experiments show that the threats can effectively bypass the restrictions and moderation policies of OpenAI, resulting in users exposing to the risk of bias, toxic content, privacy, and disinformation. We develop a lightweight, threat-agnostic defense to mitigate insider and outsider threats. Our evaluations demonstrate the efficacy of our defense.

Federated learning (FL) enables multiple participants to train a global machine learning model without sharing their private training data. Peer-to-peer (P2P) FL advances existing centralized FL paradigms by eliminating the server that aggregates local models from participants and then updates the global model. However, P2P FL is vulnerable to (i) honest-but-curious participants whose objective is to infer private training data of other participants, and (ii) Byzantine participants who can transmit arbitrarily manipulated local models to corrupt the learning process. P2P FL schemes that simultaneously guarantee Byzantine resilience and preserve privacy have been less studied. In this paper, we develop Brave, a protocol that ensures Byzantine Resilience And priVacy-prEserving property for P2P FL in the presence of both types of adversaries. We show that Brave preserves privacy by establishing that any honest-but-curious adversary cannot infer other participants' private data by observing their models. We further prove that Brave is Byzantine-resilient, which guarantees that all benign participants converge to an identical model that deviates from a global model trained without Byzantine adversaries by a bounded distance. We evaluate Brave against three state-of-the-art adversaries on a P2P FL for image classification tasks on benchmark datasets CIFAR10 and MNIST. Our results show that global models learned with Brave in the presence of adversaries achieve comparable classification accuracy to global models trained in the absence of any adversary.

Yuetai Li, Zhangchen Xu, Fengqing Jiang, Luyao Niu, Dinuka Sahabandu, Bhaskar Ramasubramanian, Radha Poovendran. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Large language models (LLMs) are shown to benefit from chain-of-thought (COT) prompting, particularly when tackling tasks that require systematic reasoning processes. On the other hand, COT prompting also poses new vulnerabilities in the form of backdoor attacks, wherein the model will output unintended malicious content under specific backdoor-triggered conditions during inference. Traditional methods for launching backdoor attacks involve either contaminating the training dataset with backdoored instances or directly manipulating the model parameters during deployment. However, these approaches are not practical for commercial LLMs that typically operate via API access. In this paper, we propose BadChain, the first backdoor attack against LLMs employing COT prompting, which does not require access to the training dataset or model parameters and imposes low computational overhead. BadChain leverages the inherent reasoning capabilities of LLMs by inserting a backdoor reasoning step into the sequence of reasoning steps of the model output, thereby altering the final response when a backdoor trigger is embedded in the query prompt. In particular, a subset of demonstrations will be manipulated to incorporate a backdoor reasoning step in COT prompting. Consequently, given any query prompt containing the backdoor trigger, the LLM will be misled to output unintended content. Empirically, we show the effectiveness of BadChain for two COT strategies across four LLMs (Llama2, GPT-3.5, PaLM2, and GPT-4) and six complex benchmark tasks encompassing arithmetic, commonsense, and symbolic reasoning. We show that the baseline backdoor attacks designed for simpler tasks such as semantic classification will fail on these complicated tasks. In addition, our findings reveal that LLMs endowed with stronger reasoning capabilities exhibit higher susceptibility to BadChain, exemplified by a high average attack success rate of 97.0\% across the six benchmark tasks on GPT-4. We also demonstrate the interpretability of BadChain by showing that the relationship between the trigger and the backdoor reasoning step can be well-explained based on the output of the backdoored model. Finally, we propose two defenses based on shuffling and demonstrate their overall ineffectiveness against BadChain. Therefore, BadChain remains a severe threat to LLMs, underscoring the urgency for the development of robust and effective future defenses.

Machine learning models that use deep neural networks (DNNs) are vulnerable to backdoor attacks. An adversary carrying out a backdoor attack embeds a predefined perturbation called a trigger into a small subset of input samples and trains the DNN such that the presence of the trigger in the input results in an adversary-desired output class. Such adversarial retraining however needs to ensure that outputs for inputs without the trigger remain unaffected and provide high classification accuracy on clean samples. Existing defenses against backdoor attacks are computationally expensive, and their success has been demonstrated primarily on image-based inputs. The increasing popularity of deploying pretrained DNNs to reduce costs of re/training large models makes defense mechanisms that aim to detect 'suspicious' input samples preferable.