Refusals must be resilient, not brittle.” Yet guarding refusals against adversarial phrasing and shifting user contexts remains difficult: large language models (LLMs) still yield to jailbreak prompts that evade safety filters and surface harmful content. We propose Refusal Activation Steering (RAS), a training-free, inference-time method that uses contrastive activations to shift LLM responses, biasing generation trajectories toward refusals without altering model weights. The approach is modular and domain-targetable, avoiding collateral refusals on benign queries while strengthening activation- space boundaries for unsafe content. On adversarial evaluations with an 8B instruction-tuned model, we find that steering improves refusal rate by ∼ 52% and reduces attack success rate by ∼ 40%, establishing a lightweight and interpretable safety layer for robust refusal consistency. To foster further research in this domain, we have made our implementation publicly available.

Alignment is crucial for text-to-image (T2I) models to ensure that the generated images faithfully capture user intent while maintaining safety and fairness. Direct Preference Optimization (DPO) has emerged as a key alignment technique for large language models (LLMs), and its influence is now extending to T2I systems. This paper introduces DPO-Kernels for T2I models, a novel extension of DPO that enhances alignment across three key dimensions: (i) Hybrid Loss, which integrates embedding-based objectives with the traditional probability-based loss to improve optimization; (ii) Kernelized Representations, leveraging Radial Basis Function (RBF), Polynomial, and Wavelet kernels to enable richer feature transformations, ensuring better separation between safe and unsafe inputs; and (iii) Divergence Selection, expanding beyond DPO’s default Kullback–Leibler (KL) regularizer by incorporating alternative divergence measures such as Wasserstein and Rényi divergences to enhance stability and robustness in alignment training. We introduce DETONATE, the first large-scale benchmark of its kind, comprising approximately 100K curated image pairs, categorized as chosen and rejected. This benchmark encapsulates three critical axes of social bias and discrimination: Race, Gender, and Disability. The prompts are sourced from the hate speech datasets, while the images are generated using state-of-the-art T2I models, including Stable Diffusion 3.5 Large (SD-3.5), Stable Diffusion XL (SD-XL), and Midjourney. Furthermore, to evaluate alignment beyond surface metrics, we introduce the Alignment Quality Index (AQI) for T2I systems: a novel geometric measure that quantifies latent space separability of safe/unsafe image activations, revealing hidden model vulnerabilities. While alignment techniques often risk overfitting, we empirically demonstrate that DPO-Kernels preserve strong generalization bounds using the theory of Heavy-Tailed Self-Regularization (HT-SR).

The rapid proliferation of Large Language Models (LLMs) has significantly contributed to the development of equitable AI systems capable of factual question-answering (QA). However, no known study tests the LLMs' robustness when presented with obfuscated versions of questions. To systematically evaluate these limitations, we propose a novel technique, ObfusQAte, and leveraging the same, introduce ObfusQA, a comprehensive, first-of-its-kind framework with multi-tiered obfuscation levels designed to examine LLM capabilities across three distinct dimensions: (i) Named-Entity Indirection, (ii) Distractor Indirection, and (iii) Contextual Overload. By capturing these fine-grained distinctions in language, ObfusQA provides a comprehensive benchmark for evaluating LLM robustness and adaptability. Our study observes that LLMs exhibit a tendency to fail or generate hallucinated responses when confronted with these increasingly nuanced variations. To foster research in this direction, we make ObfusQAte publicly available.

Abhilekh Borah, Chhavi Sharma, Danush Khanna, Utkarsh Bhatt, Gurpreet Singh, Hasnat Md Abdullah, Raghav Kaushik Ravi, Vinija Jain, Jyoti Patel, Shubham Singh, Vasu Sharma, Arpita Vats, Rahul Raja, Aman Chadha, Amitava Das. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Arijit Maji, Raghvendra Kumar, Akash Ghosh, Anushka, Nemil Shah, Abhilekh Borah, Vanshika Shah, Nishant Mishra, Sriparna Saha. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.