CoVerify Vision II establishes a credible workflow for fact-checking. It identifies high-risk content, restores degraded images, performs robust forensics, and generates interpretable evidence for human review and logging. The system supports DFD, financial fraud prevention, and media literacy. In collaboration with III and DTA, it is currently in pilot deployment with the Investigation Bureau and police forensics, serving as Taiwan’s first domestic solution to enter the public sector.

The breakthrough is a complete evidence-chain workflow for AI-generated media. The system first purifies degraded real-world inputs, then performs robust cross-compression forensic analysis, and finally converts AI signals into human-usable evidence: heatmaps, key frames, reasons, risk levels and review records. First-tier evidence includes CVPR main-conference results, IJCV work on cross-compression deepfake detection, and T-PAMI-level deepfake defense research.

Applications include fact-checking, financial anti-fraud, public-sector review, police screening, platform governance and media literacy. The team combines AI forensics, cognitive research, social-risk alerting, low-power deployment, policy collaboration and real adoption channels, turning models into executable, recordable and deployable workflows.