Maryam Saleki

Naseem Machlovi, Maryam Saleki, Ruhul Amin et al.

As large language models (LLMs) become deeply embedded in daily life, the urgent need for safer moderation systems that distinguish between naive and harmful requests while upholding appropriate censorship boundaries has never been greater. While existing LLMs can detect dangerous or unsafe content, they often struggle with nuanced cases such as implicit offensiveness, subtle gender and racial biases, and jailbreak prompts, due to the subjective and context-dependent nature of these issues. Furthermore, their heavy reliance on training data can reinforce societal biases, resulting in inconsistent and ethically problematic outputs. To address these challenges, we introduce GuardEval, a unified multi-perspective benchmark dataset designed for both training and evaluation, containing 106 fine-grained categories spanning human emotions, offensive and hateful language, gender and racial bias, and broader safety concerns. We also present GemmaGuard (GGuard), a Quantized Low-Rank Adaptation (QLoRA), fine-tuned version of Gemma3-12B trained on GuardEval, to assess content moderation with fine-grained labels. Our evaluation shows that GGuard achieves a macro F1 score of 0.832, substantially outperforming leading moderation models, including OpenAI Moderator (0.64) and Llama Guard (0.61). We show that multi-perspective, human-centered safety benchmarks are critical for mitigating inconsistent moderation decisions. GuardEval and GGuard together demonstrate that diverse, representative data materially improve safety, and adversarial robustness on complex, borderline cases.

Naseem Machlovi, Maryam Saleki, Innocent Ababio et al.

As AI systems become more integrated into daily life, the need for safer and more reliable moderation has never been greater. Large Language Models (LLMs) have demonstrated remarkable capabilities, surpassing earlier models in complexity and performance. Their evaluation across diverse tasks has consistently showcased their potential, enabling the development of adaptive and personalized agents. However, despite these advancements, LLMs remain prone to errors, particularly in areas requiring nuanced moral reasoning. They struggle with detecting implicit hate, offensive language, and gender biases due to the subjective and context-dependent nature of these issues. Moreover, their reliance on training data can inadvertently reinforce societal biases, leading to inconsistencies and ethical concerns in their outputs. To explore the limitations of LLMs in this role, we developed an experimental framework based on state-of-the-art (SOTA) models to assess human emotions and offensive behaviors. The framework introduces a unified benchmark dataset encompassing 49 distinct categories spanning the wide spectrum of human emotions, offensive and hateful text, and gender and racial biases. Furthermore, we introduced SafePhi, a QLoRA fine-tuned version of Phi-4, adapting diverse ethical contexts and outperforming benchmark moderators by achieving a Macro F1 score of 0.89, where OpenAI Moderator and Llama Guard score 0.77 and 0.74, respectively. This research also highlights the critical domains where LLM moderators consistently underperformed, pressing the need to incorporate more heterogeneous and representative data with human-in-the-loop, for better model robustness and explainability.

Murat Sensoy, Lance M. Kaplan, Simon Julier et al.

Autonomous and semi-autonomous systems are using deep learning models to improve decision-making. However, deep classifiers can be overly confident in their incorrect predictions, a major issue especially in safety-critical domains. The present study introduces three foundational desiderata for developing real-world risk-aware classification systems. Expanding upon the previously proposed Evidential Deep Learning (EDL), we demonstrate the unity between these principles and EDL's operational attributes. We then augment EDL empowering autonomous agents to exercise discretion during structured decision-making when uncertainty and risks are inherent. We rigorously examine empirical scenarios to substantiate these theoretical innovations. In contrast to existing risk-aware classifiers, our proposed methodologies consistently exhibit superior performance, underscoring their transformative potential in risk-conscious classification strategies.

Murat Sensoy, Lance Kaplan, Federico Cerutti et al.

Deep neural networks are often ignorant about what they do not know and overconfident when they make uninformed predictions. Some recent approaches quantify classification uncertainty directly by training the model to output high uncertainty for the data samples close to class boundaries or from the outside of the training distribution. These approaches use an auxiliary data set during training to represent out-of-distribution samples. However, selection or creation of such an auxiliary data set is non-trivial, especially for high dimensional data such as images. In this work we develop a novel neural network model that is able to express both aleatoric and epistemic uncertainty to distinguish decision boundary and out-of-distribution regions of the feature space. To this end, variational autoencoders and generative adversarial networks are incorporated to automatically generate out-of-distribution exemplars for training. Through extensive analysis, we demonstrate that the proposed approach provides better estimates of uncertainty for in- and out-of-distribution samples, and adversarial examples on well-known data sets against state-of-the-art approaches including recent Bayesian approaches for neural networks and anomaly detection methods.