Publication Type

Conference Proceeding Article

Version

publishedVersion

Publication Date

4-2026

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their deployment is frequently undermined by undesirable behaviors such as generating harmful content, factual inaccuracies, and societal biases. Diagnosing the root causes of these failures poses a critical challenge for AI safety. Existing attribution methods, particularly those based on parameter gradients, often fall short due to prohibitive noisy signals and computational complexity. In this work, we introduce a novel and efficient framework that diagnoses a range of undesirable LLM behaviors by analyzing representation and its gradients, which operates directly in the model's activation space to provide a semantically meaningful signal linking outputs to their training data. We systematically evaluate our method for tasks that include tracking harmful content, detecting backdoor poisoning, and identifying knowledge contamination. The results demonstrate that our approach not only excels at sample-level attribution but also enables fine-grained token-level analysis, precisely identifying the specific samples and phrases that causally influence model behavior. This work provides a powerful diagnostic tool to understand, audit, and ultimately mitigate the risks associated with LLMs.

Discipline

Artificial Intelligence and Robotics | Information Security

Research Areas

Software and Cyber-Physical Systems

Areas of Excellence

Digital transformation

Publication

Proceedings of the Fourteenth International Conference on Learning Representations (ICLR 2026), Rio de Janeiro, Brazil, April 23-27

First Page

1

Last Page

15

Publisher

ICLR

City or Country

Rio de Janeiro, Brazil

Additional URL

https://openreview.net/forum?id=MN1qlAVJLV

Download

Share

COinS