cs.CRNov 18, 2026

Reject and Report: Forgery-Evident Undetectable Watermarks for Generative Image Models

Authors: Yihan Wang, Haochen Sun, Xi He, Yaoliang Yu, N. Asokan

Organizations: University of Waterloo · Vector Institute · KTH Royal Institute of Technology

Abstract

Provenance schemes such as those based on content watermarks let images from a generative model be traced back to it. A practical watermarking scheme must be accurate at a low false-positive rate (FPR), robust under content-preserving processing, and undetectable. In addition, it must resist forgery; otherwise, even from a single watermarked image and a surrogate model, an adversary can imprint the watermark onto an arbitrary image or regenerate new content that carries it, falsely attributing harmful content to a model or making authentic photographs appear AI-generated. We argue that a watermarking scheme should not only reject forgeries but also report them: a forgery rejected as unwatermarked leaves no trace of the attempt. State-of-the-art forgery-resilient undetectable watermarking schemes, such as SemBind, fall short on both counts. External anomaly detectors that do report forgeries either miss most of them or falsely flag genuine images after common processing. We propose Tripwire-WM, the first forgery-evident framework for undetectable watermarks. It commits the semantics of the generated image to the magnitudes of the initial latent noise, which the sign-based watermarks leave unused, through a distribution-preserving embedding; a forged image carries a valid watermark but with a commitment that does not match its content, so the detector both rejects it and reports it as forged. Built on the pseudorandom code (PRC) watermark and Gaussian Shading++, Tripwire-WM provably preserves their undetectability. On MS-COCO, with the owner's own model as the surrogate, it accepts no imprinting forgery, compared with up to 55.0% for SemBind, and reports all of them as forged; it also accepts at most 1.8% of reprompting forgeries, against up to 46.2% for SemBind. Meanwhile, it accepts at least 99% of genuine watermarked images at 10^{-6} FPR and matches the robustness of the base schemes.

Figures & tables

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Aug 4, 2026

Attribute-based Undetectable Watermarking for Generative AI Models

Generative AI systems increasingly produce content whose provenance is difficult to verify, motivating watermarking techniques for identifying model-generated outputs. Existing cryptographic watermarking methods provide strong undetectability guarantees: without a detection key, watermarked outputs are computationally indistinguishable from unwatermarked ones. However, these approaches do not address the crucial deployment challenge of how to safely delegate detection capabilities. With an unrestricted detection key, a malicious detector may use the detection key beyond its intended scope, enabling watermark sanitization, scope abuse, and user profiling. To mitigate this safety concern, we introduce, to the best of our knowledge, the first attribute-based watermarking for generative AI models, providing fine-grained, policy-controlled watermark detection. In our approach, each generated output is associated with attributes, and each detection key is constrained by a policy on potential attributes. A detection key can only be used to detect watermarked outputs whose attributes satisfy the corresponding policy, while watermarked outputs that fall outside the policy remain computationally indistinguishable from unwatermarked ones. We construct such an attribute-based watermarking scheme and formalize its security properties, including consistency, adaptive robustness to bounded corruptions, undetectability, and soundness, along with a security proof under standard cryptographic assumptions. Our construction integrates constrained pseudorandom functions, pseudorandom error-correcting codes, and randomness recovery procedures with generative AI models. Finally, we implement a prototype and an empirical evaluation, demonstrating that attribute-based watermarking is both effective and practical.
May 2, 2026

CSGuard: Toward Forgery-Resistant Watermarking in Diffusion Models via Compressed Sensing Constraint

Latent-based diffusion model watermarking embeds watermarks into generated images' latent space to enable content attribution, offering a training-free solution for intellectual property protection and digital forensics. However, these methods exhibit a critical vulnerability to the forgery attack, attackers can extract the watermark by inverting the watermarked image and re-generating it with an arbitrary prompt, thereby enabling false attribution on malicious content. In this paper, we propose the CSGuard, the first forgery-resistant watermarking schema that leverages compressed sensing to bind the watermarked image generation and verification to a secret matrix. This ensures that only users possessing the secret matrix can correctly embed or verify the image watermark, prevents the illegal users from forgery without compromising generation quality and watermark integrity. Experimental results demonstrate that CSGuard achieves strong forgery resistance, reduces the attack success rate from 100.0% to 28.12%, and achieve 100% detection rate on benign watermarked images without compromising watermarking effectiveness.
Apr 13, 2026

On the Robustness of Watermarking for Autoregressive Image Generation

The proliferation of autoregressive (AR) image generators demands reliable detection and attribution of their outputs to mitigate misinformation, and to filter synthetic images from training data to prevent model collapse. To address this need, watermarking techniques, specifically designed for AR models, embed a subtle signal at generation time, enabling downstream verification through a corresponding watermark detector. In this work, we study these schemes and demonstrate their vulnerability to both watermark removal and forgery attacks. We assess existing attacks and further introduce three new attacks: (i) a vector-quantized regeneration removal attack, (ii) adversarial optimization-based attack, and (iii) a frequency injection attack. Our evaluation reveals that removal and forgery attacks can be effective with access to a single watermarked reference image and without access to original model parameters or watermarking secrets. Our findings indicate that existing watermarking schemes for AR image generation do not reliably support synthetic content detection for dataset filtering. Moreover, they enable Watermark Mimicry, whereby authentic images can be manipulated to imitate a generator's watermark and trigger false detection to prevent their inclusion in future model training.