Research map / Video / Video-language understanding
SafeGuard: A Multi-Agent Perception-Reasoning Framework for Social-Risk AI-Generated Video Detection
Wenlin Wu, Sheng Zhou, Peipei Song, Wenhao Wang, Junbin Xiao, Xun Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add social risk-aware hypothesis generation and routing (SHGR) to GPT-4o (ablation) | GPT-4o baseline · Overall Avg | accuracy (%): 66.8 -> 68.5 | Improved |
| add video cropping (VC) after SHGR (ablation) | GPT-4o + SHGR · Overall Avg | accuracy (%): 68.5 -> 72.4 | Improved |
| add forensic trace extraction (FTE) after SHGR+VC (ablation) | GPT-4o + SHGR + VC · Overall Avg | accuracy (%): 72.4 -> 84.1 | Improved |
| add self-reflective verifier (SRV) to SHGR+VC+FTE (ablation) | GPT-4o + SHGR + VC + FTE · Overall Avg | accuracy (%): 84.1 -> 89.3 | Improved |
| add SRV without FTE (SHGR+VC) (ablation) | GPT-4o + SHGR + VC · Overall Avg | accuracy (%): 72.4 -> 81.8 | Improved |
| add SRV without FTE (SHGR+VC) on SafeVid F1 is a minor gain (ablation) | GPT-4o + SHGR + VC · SafeVid Avg | F1 (%): 71.0 -> 71.8 | Improved |
| add SHGR alone on SafeVid F1 (ablation) | GPT-4o baseline · SafeVid Avg | F1 (%): 60.5 -> 59.1 | Hurt |
| add optical-flow tool to SRV-only variant (w/o tool use) (ablation) | SafeGuard · Overall Avg | accuracy (%): 81.8 -> 83.0 | Improved |
| add depth tool to optical flow (ablation) | SafeGuard · Overall Avg | accuracy (%): 83.0 -> 85.0 | Improved |
| add appearance tool (ablation) | SafeGuard · Overall Avg | accuracy (%): 85.0 -> 88.0 | Improved |
| add pixel-level tool (ablation) | SafeGuard · Overall Avg | accuracy (%): 88.0 -> 89.3 | Improved |
| use SafeGuard over plain GPT-4o on out-of-domain SafeVid | GPT-4o · SafeVid OOD | accuracy (%): 63.7 -> 80.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BusterX++ — BusterX: MLLM-Powered AI-Generated Video Forgery Detection and Explanation (NeurIPS 2026) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Locate-Then-Examine: Grounded Region Reasoning Improves Detection of AI-Generated Images (CVPR 2026 · similarity 0.87)
- Seeing What Matters: Generalizable AI-generated Video Detection with Forensic-Oriented Augmentation (NeurIPS 2025 · similarity 0.86)
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.86)
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.86)
- ReAgent-V: A Reward-Driven Multi-Agent Framework for Video Understanding (NeurIPS 2025 · similarity 0.86)
- VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
