DeepForgeSeal: Latent Space-Driven Semi-Fragile Watermarking for Deepfake Detection Using Multi-Agent Adversarial Reinforcement Learning

2025-11-07

AI Summary

To address the poor generalizability of passive deepfake detection methods and the fundamental trade-off in existing watermarking schemes—namely, the difficulty in simultaneously achieving robustness against benign distortions and sensitivity to malicious manipulations—this paper proposes a latent-space-based semi-fragile watermarking framework. Methodologically, it innovatively integrates latent encoding from deep generative models, differentiable watermark embedding/extraction networks, and multi-agent adversarial reinforcement learning (MARL), where collaborative agents dynamically simulate diverse image transformations to optimize the watermark’s equilibrium between robustness and fragility. Its key contribution is the first application of MARL to semi-fragile watermark design, enabling active, adaptive synthetic media identification. Evaluated on CelebA and CelebA-HQ, the framework achieves detection accuracy improvements of 4.5% and 5.3%, respectively, significantly outperforming state-of-the-art approaches.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Rapid advances in generative AI have led to increasingly realistic deepfakes, posing growing challenges for law enforcement and public trust. Existing passive deepfake detectors struggle to keep pace, largely due to their dependence on specific forgery artifacts, which limits their ability to generalize to new deepfake types. Proactive deepfake detection using watermarks has emerged to address the challenge of identifying high-quality synthetic media. However, these methods often struggle to balance robustness against benign distortions with sensitivity to malicious tampering. This paper introduces a novel deep learning framework that harnesses high-dimensional latent space representations and the Multi-Agent Adversarial Reinforcement Learning (MAARL) paradigm to develop a robust and adaptive watermarking approach. Specifically, we develop a learnable watermark embedder that operates in the latent space, capturing high-level image semantics, while offering precise control over message encoding and extraction. The MAARL paradigm empowers the learnable watermarking agent to pursue an optimal balance between robustness and fragility by interacting with a dynamic curriculum of benign and malicious image manipulations simulated by an adversarial attacker agent. Comprehensive evaluations on the CelebA and CelebA-HQ benchmarks reveal that our method consistently outperforms state-of-the-art approaches, achieving improvements of over 4.5% on CelebA and more than 5.3% on CelebA-HQ under challenging manipulation scenarios.

Problem

• Balancing robustness and fragility in deepfake watermarking systems
• Generalizing detection capabilities across evolving deepfake techniques
• Developing adaptive watermarking resistant to manipulation while sensitive to tampering

Innovation

• Uses latent space representations for watermark embedding
• Employs multi-agent adversarial reinforcement learning paradigm
• Balances robustness and fragility via dynamic curriculum learning