Synthetic & Adversarial ForEnsics
(SAFE)

Topics: Synthetic media detection, adversarial attacks, multimedia forensics, robustness, AI security, generative models

We invite original research contributions in (but not limited to) the following areas:

• Generative image and video models (e.g., diffusion models, VLMs, GANs) and foundation models for synthetic media creation, manipulation, and robustness evaluation
• Defensive fingerprinting and watermarking of generative models for source attribution
• Red teaming methods for generative models, including realistic adversarial simulation environments
• Provenance tracking and authenticity verification of digital media
• Detection of synthetic media (e.g., AI-generated images, and video)
• Generation and detection of synthetic multimodal data (e.g., image and audio)
• Adversarial attack, detection, and mitigation methods for visual data including video streams
• Training data poisoning detection and mitigation for generative models
• Joint modeling of synthetic media and adversarial threats
• Benchmarks, datasets, and evaluation metrics for adversarial detection and robustness
• Generative and foundation model explanation and interpretation
• Adversarial attack and mitigation techniques for distilled models
• Ethical, societal, and legal implications of synthetic and adversarial media

Accepted papers will be presented as posters. Full papers (6–8 pages, excl. refs) are archival and included in the CVPR 2026 Workshop Proceedings; short papers and extended abstracts (2–4 pages, excl. refs) are not. All submissions undergo double-blind peer review and must follow the appropriate CVPR 2026 formatting guidelines.