Watermarking & Deepfake Detection
Cross-Paradigm Robustness Benchmarking for Deepfake Detectors
Generated automatically from the limitations stated in 3 papers (ICML, NeurIPS, CVPR), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.
The problem
Current deepfake detectors are developed and validated within narrow manipulation regimes, such as full-frame image synthesis, boundary-blended face swaps, or identity-conditioned video pairs. Because detection models rely on artifacts specific to their target generation family, their performance characteristics across alternative forgery paradigms—such as expression reenactment, localized facial attribute edits, and full-image generation—remain untested. Consequently, practitioners cannot determine whether existing detectors offer any protection outside their specific training domain or if they fail completely when deployed against unmodeled manipulation types.
Why it matters
Unlocks an empirical map of cross-forgery failure modes, establishing which manipulation types transfer across detector paradigms and which require distinct detection techniques.
Ways to approach it
Prior-work checks are free with an account. Results someone already ran are shown to everyone.
- 1
Standardized Multi-Paradigm Benchmark: Assemble an evaluation testbed pairing leading detectors (spatial-artifact, frequency-based, and identity-contrastive models) against standardized suites of full-face synthesis, expression swaps (e.g., Face2Face, NeuralTextures), localized attribute edits, and modern diffusion-based face swaps, measuring cross-manipulation AUC-ROC and false-positive degradation.
- 2
Training Regime Sensitivity Audit: Train representative baseline detectors under isolated manipulation assumptions (e.g., identity contrast pairs vs. whole-image blending masks) and measure how detector transferability drops as the testing manipulation diverges in spatial extent and semantic alteration.
- 3
Feature Representation Probing: Extract intermediate detector embeddings across different forgery mechanisms to quantify whether detectors learn domain-general forgery cues or merely overfit to pipeline-specific synthesis artifacts.
Have a different approach?
Describe how you would tackle this problem and we'll look for papers that already do it. Free; your text stays private.
Why it might fail
If concurrent comprehensive benchmark papers evaluate all modern detector families across these exact forgery types, or if standard foundation model features already yield near-perfect zero-shot cross-manipulation generalization across all categories.
Evidence
Each paper's own statement of the limitation, verbatim.
- Fleet: Few Shots Lead Effective AI-generated Image DetectionICML 2026
Restricted to full-image generation; generalization to localized editing or face-swapping is untested
- FreqBlender: Enhancing DeepFake Detection by Blending Frequency KnowledgeNeurIPS 2024
Inherits the face-swapping assumption of spatial blending methods; unvalidated on whole-face synthesis or attribute editing forgeries
- Implicit Identity Driven Deepfake Face Swapping DetectionCVPR 2023
Requires knowledge of target-face identities and source-target-fake video triples during training, and does not apply to expression-swap methods (Face2Face, NeuralTextures) where the identity contrast is inapplicable, so those samples only use IIE constraints
Nearest existing work
- DeepProtect: Proactive Face-Swapping Defense using Identity Blending and Attribute DistortionCVPR 2026
- DeeperForensics-1.0: A Large-Scale Dataset for Real-World Face Forgery DetectionCVPR 2020
- FaceForensics++: Learning to Detect Manipulated Facial ImagesICCV 2019
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter TuningCVPR 2025
- OpenForensics: Large-Scale Challenging Dataset for Multi-Face Forgery Detection and Segmentation In-the-WildICCV 2021
- FaceShield: Defending Facial Image against Deepfake ThreatsICCV 2025
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face DeepfakesNeurIPS 2025
- Celeb-DF: A Large-Scale Challenging Dataset for DeepFake ForensicsCVPR 2020
- Transcending Forgery Specificity with Latent Space Augmentation for Generalizable Deepfake DetectionCVPR 2024
- DeepFake Disrupter: The Detector of DeepFake Is My FriendCVPR 2022
- DeepShield: Fortifying Deepfake Video Detection with Local and Global Forgery AnalysisICCV 2025
- Advancing High Fidelity Identity Swapping for Forgery DetectionCVPR 2020
- Detecting Deepfakes With Self-Blended ImagesCVPR 2022
- Proactive Defense Benchmark against Deepfake GenerationICML 2026
- NullSwap: Proactive Identity Cloaking Against Deepfake Face SwappingICCV 2025
Related open problems
Watermarking & Deepfake Detection
Black-Box Verification and Trigger Generation for Model Watermarking and Detection
Current watermarking and deepfake detection methods require full white-box access to target model parameters and gradients during trigger construction or verification. When proprietary models are deployed solely behind inference APIs or distributed as encrypted binaries, these techniques cannot be applied at all. This leaves model owners and auditors unable to verify intellectual property theft, track provenance, or detect misuse across commercial black-box deployments.
Watermarking & Deepfake Detection
Black-Box and Model-Agnostic Deepfake and Watermark Verification for Closed-Source Generative Models
Current watermark verification and deepfake detection methods require white-box access to target model parameters, intermediate layer activations, or gradient information. Because leading generative models and vision-language systems are served exclusively behind closed commercial APIs, internal inspection methods cannot be executed by downstream verifiers or auditors. Consequently, defenders cannot detect deepfakes or verify watermarks when the generating model is proprietary, unknown, or inaccessible. Removing the precondition of internal weight and activation access is necessary for auditing real-world deployed models.
Watermarking & Deepfake Detection
Black-Box and Architecture-Agnostic Watermarking and Defense for Diffusion Models
Current proactive watermarking, attribution, and anti-customization defenses structurally assume white-box access to target latent diffusion models, requiring direct inspection of latent spaces, VAE fine-tuning, or gradient backpropagation through internal U-Net attention layers. In practical deployment scenarios, defenders and copyright holders must protect content against proprietary black-box APIs, pixel-space diffusion models, or non-VAE architectures where internal activations and weights are inaccessible. Because existing methods are tethered to specific LDM preconditions, there is currently no verified mechanism to provide provenance or anti-fine-tuning protection across arbitrary or black-box diffusion pipelines.
Watermarking & Deepfake Detection
Cross-Family Generalization and Robustness of Text Watermarking Attacks and Defenses
Existing evaluations of text watermark attacks, detection, and removal methods have almost exclusively tested narrow subsets of logit-based schemes (e.g., KGW and SIR) evaluated on fixed base models like OPT-1.3B. Because no single study has evaluated these attack and defense methods across fundamentally different watermark families—including sampling-based, dynamic-hash, content-adaptive, and error-correction-augmented schemes—practitioners cannot determine whether reported evasion and detection results reflect general properties or family-specific artifacts. Consequently, deployment decisions for provenance tracking in production language models rely on unverified theoretical compatibility claims rather than measured empirical robustness.