PatchBench: Measuring Collateral Damage in Activation Patching

arXiv:2610.10276 · cs.LG, cs.CL, cs.CR · Submitted 2026-10-07 · Read on arXiv

cs.LG, cs.CL, cs.CR

Submitted: 2026-10-07

Updated: 2026-10-07

Comments: Accepted to NeurIPS 2026 (Datasets and Benchmarks Track)

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Activation Patching — 1× in this paper · explained in 2 episodes

Sources

• Refusal in Language Models Is Mediated by a Single Direction
• Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions
• Are aligned neural networks adversarially aligned?
• JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models
• RedBench: A Universal Dataset for Comprehensive Red Teaming of Large Language Models
• WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs
• Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations
• Activation Surgery: Jailbreaking White-box LLMs without Touching the Prompt
• Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategy
• WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Models
• Programming Refusal with Conditional Activation Steering
• Inference-Time Intervention: Eliciting Truthful Answers from a Language Model
• JailBreakV: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks
• Training language models to follow instructions with human feedback
• Safety Alignment Should Be Made More Than Just a Few Tokens Deep
• GRP-Obliteration: Unaligning LLMs With a Single Unlabeled Prompt
• Large Language Model Alignment: A Survey
• "Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models
• AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint
• Llama 2: Open Foundation and Fine-Tuned Chat Models

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks