20 March 2026 in ACM Transactions on Intelligent Systems and Technology
As Deep Neural Networks (DNNs) become integral to critical applications, protecting their Intellectual Property (IP) has become paramount. Neural network watermarking is a technique that embeds unique identifiers into models, asserting ownership and deterring unauthorized use. However, sophisticated attacks can deactivate or remove these watermarks without significantly compromising model performance, undermining current protection strategies. In this article, we introduce the first method for reactivating deactivated neural network watermarks in altered DNN models without requiring access to the original model parameters or training data. By formulating the reactivation process as an optimization problem, we employ projected gradient descent to identify new trigger inputs that restore the embedded watermark. Regularization techniques are incorporated to ensure these triggers resemble legitimate inputs, enhancing both stealth and practicality. Through experiments on various benchmark datasets and model architectures, we demonstrate the effectiveness of our method against common model alterations, including fine-tuning, pruning, and surrogate model attacks. Our work addresses a critical gap in DNN IP protection, offering a robust and practical solution for watermark reactivation. This empowers model owners to assert their rights even in the face of advanced adversarial tactics.
1 January 2026 in IEEE Transactions on Network Science and Engineering
11 June 2026 in ACM Transactions on Privacy and Security
Deep neural networks for image classification require protection against unauthorized use and redistribution. Existing watermarking methods suffer from a critical vulnerability: watermarks are always active and detectable, allowing adversaries to identify and remove them before deployment. We propose DormMark, a novel framework for image classification models that introduces delayed-activation watermarks which remain dormant and hidden under deployment-time black-box query auditing during initial deployment, but automatically activate upon fine-tuning. Our approach employs a three-stage training paradigm: (1) embedding watermarks using triggered samples, (2) masking to suppress watermark functionality while preserving its latent presence, and (3) activation through standard fine-tuning without owner intervention. This mechanism exploits neural networks’ forgetting-remembering behaviors during continued training, creating a fragile equilibrium that behaves similarly to a clean model under deployment-time black-box auditing but reliably manifests ownership indicators after modification. We consider a private-key black-box verification setting in which the owner keeps the concrete trigger instances secret. Experiments across multiple architectures (VGG19, ResNet-18/56, DenseNet-121, WideResNet-34) and datasets (CIFAR-10, CIFAR-100, GTSRB) demonstrate 100% watermark success rates, high imperceptibility (PSNR > 38 dB, SSIM = 0.99), negligible accuracy loss (< 0.04%), and robustness against 80% parameter pruning. DormMark represents a paradigm shift from static to conditionally-activated ownership verification, providing a more robust framework for intellectual property protection.