The Rogue Scalpel: Activation Steering Compromises LLM Safety

Activation steering is a promising technique for controlling LLM behavior by adding semantically meaningful vectors directly into a model's hidden states during inference. It is often framed as a precise, interpretable, and potentially safer alternative to fine-tuning. We demonstrate the opposite: steering systematically breaks model alignment safeguards, making it comply with harmful requests. Through extensive experiments on different model families, we show that even steering in a random direction can increase the probability of harmful compliance from 0% to 2-27%. Alarmingly, steering benign features from a sparse autoencoder (SAE), a common source of interpretable directions, increases these rates by a further 2-4%. Finally, we show that combining 20 randomly sampled vectors that jailbreak a single prompt creates a universal attack, significantly increasing harmful compliance on unseen requests. These results challenge the paradigm of safety through interpretability, showing that precise control over model internals does not guarantee precise control over model behavior.

Lightweight error mitigation strategies for post-training N:M activation sparsity in LLMs

The demand for efficient large language model (LLM) inference has intensified the focus on sparsification techniques. While semi-structured (N:M) pruning is well-established for weights, its application to activation pruning remains underexplored despite its potential for dynamic, input-adaptive compression and reductions in I/O overhead. This work presents a comprehensive analysis of methods for post-training N:M activation pruning in LLMs. Across multiple LLMs, we demonstrate that pruning activations enables superior preservation of generative capabilities compared to weight pruning at equivalent sparsity levels. We evaluate lightweight, plug-and-play error mitigation techniques and pruning criteria, establishing strong hardware-friendly baselines that require minimal calibration. Furthermore, we explore sparsity patterns beyond NVIDIA's standard 2:4, showing that the 16:32 pattern achieves performance nearly on par with unstructured sparsity. However, considering the trade-off between flexibility and hardware implementation complexity, we focus on the 8:16 pattern as a superior candidate. Our findings provide both effective practical methods for activation pruning and a motivation for future hardware to support more flexible sparsity patterns. Our code is available https://anonymous.4open.science/r/Structured-Sparse-Activations-Inference-EC3C/README.md .

OrtSAE: Orthogonal Sparse Autoencoders Uncover Atomic Features

Sparse autoencoders (SAEs) are a technique for sparse decomposition of neural network activations into human-interpretable features. However, current SAEs suffer from feature absorption, where specialized features capture instances of general features creating representation holes, and feature composition, where independent features merge into composite representations. In this work, we introduce Orthogonal SAE (OrtSAE), a novel approach aimed to mitigate these issues by enforcing orthogonality between the learned features. By implementing a new training procedure that penalizes high pairwise cosine similarity between SAE features, OrtSAE promotes the development of disentangled features while scaling linearly with the SAE size, avoiding significant computational overhead. We train OrtSAE across different models and layers and compare it with other methods. We find that OrtSAE discovers 9% more distinct features, reduces feature absorption (by 65%) and composition (by 15%), improves performance on spurious correlation removal (+6%), and achieves on-par performance for other downstream tasks compared to traditional SAEs.

I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders

Large Language Models (LLMs) have achieved remarkable success in natural language processing. Recent advances have led to the developing of a new class of reasoning LLMs; for example, open-source DeepSeek-R1 has achieved state-of-the-art performance by integrating deep thinking and complex reasoning. Despite these impressive capabilities, the internal reasoning mechanisms of such models remain unexplored. In this work, we employ Sparse Autoencoders (SAEs), a method to learn a sparse decomposition of latent representations of a neural network into interpretable features, to identify features that drive reasoning in the DeepSeek-R1 series of models. First, we propose an approach to extract candidate ''reasoning features'' from SAE representations. We validate these features through empirical analysis and interpretability methods, demonstrating their direct correlation with the model's reasoning abilities. Crucially, we demonstrate that steering these features systematically enhances reasoning performance, offering the first mechanistic account of reasoning in LLMs. Code available at https://github.com/AIRI-Institute/SAE-Reasoning

CLEAR: Character Unlearning in Textual and Visual Modalities

Machine Unlearning (MU) is critical for enhancing privacy and security in deep learning models, particularly in large multimodal language models (MLLMs), by removing specific private or hazardous information. While MU has made significant progress in textual and visual modalities, multimodal unlearning (MMU) remains significantly underexplored, partially due to the absence of a suitable open-source benchmark. To address this, we introduce CLEAR, a new benchmark designed to evaluate MMU methods. CLEAR contains 200 fictitious individuals and 3,700 images linked with corresponding question-answer pairs, enabling a thorough evaluation across modalities. We assess 10 MU methods, adapting them for MMU, and highlight new challenges specific to multimodal forgetting. We also demonstrate that simple $\ell_1$ regularization on LoRA weights significantly mitigates catastrophic forgetting, preserving model performance on retained data. The dataset is available at https://huggingface.co/datasets/therem/CLEAR