Neural Honeytrace: Plug&Play Watermarking Framework against Model Extraction Attacks

Triggerable watermarking enables model owners to assert ownership against model extraction attacks. However, most existing approaches require additional training, which limits post-deployment flexibility, and the lack of clear theoretical foundations makes them vulnerable to adaptive attacks. In this paper, we propose Neural Honeytrace, a plug-and-play watermarking framework that operates without retraining. We redefine the watermark transmission mechanism from an information perspective, designing a training-free multi-step transmission strategy that leverages the long-tailed effect of backdoor learning to achieve efficient and robust watermark embedding. Extensive experiments demonstrate that Neural Honeytrace reduces the average number of queries required for a worst-case t-test-based ownership verification to as low as 2% of existing methods, while incurring zero training cost.

Neural Honeytrace is a new watermarking method that helps AI model owners prove ownership after their models are copied through repeated queries. Unlike existing approaches, it does not require retraining the model, making it easier to deploy in real-world systems. Experiments show that Neural Honeytrace can reduce the number of queries needed for ownership verification to as low as 2% of existing methods while introducing no additional training cost.