AI Insight
This paper introduces the first fingerprinting methods for multimodal large language models (MLLMs) to protect against unauthorized copying and distillation. The researchers developed two complementary techniques: AttnPrint, which extracts low-frequency components from cross-modal attention patterns as unique fingerprints for white-box verification, and DistillTrace, which uses hypothesis testing on model outputs for black-box detection. Testing across 154 model instances and 19 architectures demonstrated that AttnPrint successfully detects derivative models while remaining robust to various modification techniques, and DistillTrace can identify distillation relationships without requiring access to model parameters.
Why it matters
As MLLMs become more valuable and widely deployed, protecting intellectual property and preventing unauthorized use becomes critical for AI companies and researchers. These fingerprinting methods provide practical tools for detecting model theft and unauthorized distillation, which could help establish accountability in AI development and deployment.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: While multimodal large language models (MLLMs) enable a wide range of image-text reasoning tasks, recent incidents indicate that they are vulnerable to illicit deployment and unauthorized distillation. Existing solutions for model provenance are typically confounded by shared language backbones in MLLMs and struggle to detect violations of distillation. To bridge this gap and safeguard model ownership, we present the first study on multimodal model fingerprinting. Inspired by recent findings that self-attention acts as a low-pass filter and that its low-frequency components are informative, we develop AttnPrint for white-box provenance. Specifically, we extract cross-modal attention distributions and isolate their low-frequency components to serve as model fingerprints. To facilitate black-box auditing, we further introduce DistillTrace, which employs hypothesis testing of MLLM outputs to identify potential model infringement. We conduct extensive experiments on 154 model instances across 19 multimodal architectures. Notably, AttnPrint achieves strong derivative-model detection performance while remaining robust to five downstream modification techniques. DistillTrace also provides evidence of distillation relationships under three parameter-independent techniques.