Large Language Models (LLMs) exhibit impressive capabilities, but remain susceptible to a growing spectrum of safety risks, including jailbreaks, toxic content, hallucinations, and bias. Existing defenses often address only a single threat type or resort to rigid outright rejection, sacrificing user experience and failing to generalize across diverse and novel attacks. This paper introduces Adversarial Scenario Extrapolation (ASE), a novel inference-time computation framework that leverages Chain-of-Thought (CoT) reasoning to simultaneously enhance LLM robustness and seamlessness. ASE guides the LLM through a self-generative process of contemplating potential adversarial scenarios and formulating defensive strategies before generating a response to the user query. Comprehensive evaluation on four adversarial benchmarks with four latest LLMs shows that ASE achieves near-zero jailbreak attack success rates and minimal toxicity, while slashing outright rejections to

Federated learning (FL) has become a key component in various language modeling applications such as machine translation, next-word prediction, and medical record analysis. These applications are trained on datasets from many FL participants that often include privacy-sensitive data, such as healthcare records, phone/credit card numbers, login credentials, etc. Although FL enables computation without necessitating clients to share their raw data, existing works show that privacy leakage is still probable in federated language models. In this paper, we present two novel findings on the leakage of privacy-sensitive user data from federated large language models without requiring access to gradients. Firstly, we make a key observation that model snapshots from the intermediate rounds in FL can cause greater privacy leakage than the final trained model. Secondly, we identify that a malicious FL participant can aggravate the leakage by tampering with the model's selective weights that are responsible for memorizing the sensitive training data of some other clients, even without any cooperation from the server. Our best-performing method increases the membership inference recall by 29% and achieves up to 71% private data reconstruction, evidently outperforming existing attacks that consider much stronger adversary capabilities. Lastly, we recommend a balanced suite of techniques for an FL client to defend against such privacy risk.

Fine-tuning large language models on private data for downstream applications poses significant privacy risks in potentially exposing sensitive information. Several popular community platforms now offer convenient distribution of a large variety of pre-trained models, allowing anyone to publish without rigorous verification. This scenario creates a privacy threat, as pre-trained models can be intentionally crafted to compromise the privacy of fine-tuning datasets. In this study, we introduce a novel poisoning technique that uses model-unlearning as an attack tool. This approach manipulates a pre-trained language model to increase the leakage of private data during the fine-tuning process. Our method enhances both membership inference and data extraction attacks while preserving model utility. Experimental results across different models, datasets, and fine-tuning setups demonstrate that our attacks significantly surpass baseline performance. This work serves as a cautionary note for users who download pretrained models from unverified sources, highlighting the potential risks involved.

Federated learning (FL) is revolutionizing how we learn from data. With its growing popularity, it is now being used in many safety-critical domains such as autonomous vehicles and healthcare. Since thousands of participants can contribute in this collaborative setting, it is, however, challenging to ensure security and reliability of such systems. This highlights the need to design FL systems that are secure and robust against malicious participants’ actions while also ensuring high utility, privacy of local data, and efficiency. In this paper, we propose a novel FL framework dubbed as FLShield that utilizes benign data from FL participants to validate the local models before taking them into account for generating the global model. This is in stark contrast with existing defenses relying on server’s access to clean datasets—an assumption often impractical in real-life scenarios and conflicting with the fundamentals of FL. We conduct extensive experiments to evaluate our FLShield framework in different settings and demonstrate its effectiveness in thwarting various types of poisoning and backdoor attacks including a defense-aware one. FLShield also preserves privacy of local data against gradient inversion attacks.

This article addresses the class imbalance issue in a low-resource language called Bengali. As a use-case, we choose one of the most fundamental NLP tasks, i.e., text classification, where we utilize three benchmark text corpora: fake-news dataset, sentiment analysis dataset, and song lyrics dataset. Each of them contains a critical class imbalance. We attempt to tackle the problem by applying several strategies that include data augmentation with synthetic samples via text and embedding generation in order to augment the proportion of the minority samples. Moreover, we apply ensembling of deep learning models by subsetting the majority samples. Additionally, we enforce the focal loss function for class-imbalanced data classification. We also apply the outlier detection technique, data resampling, and hidden feature extraction to improve the minority-f1 score. All of our experimentations are entirely focused on textual content analysis, which results in a more than90%minority f1 score for each of the three tasks. It is an excellent outcome on such highly class-imbalanced datasets.

Despite being the sixth most widely spoken language in the world, Bangla has barely received any attention in the domain of audio-visual news classification. In this work, we collect, annotate, and prepare a comprehensive news audio dataset in Bangla, comprising 5120 news clips, with around 820 hours of total duration. We also conduct practical experiments to obtain a human baseline for the news audio classification task. Later, we implement one of the human approaches by performing news classification directly on the audio features using various state-of-the-art classifiers and a few transfer learning models. To the best of our knowledge, this is the very first work developing a benchmark dataset for news audio classification in Bangla.