When a video arrives purporting to show a world leader making a controversial statement, you need to know within minutes whether the footage is genuine. We evaluated nine deepfake detection tools against 200 videos: 100 authentic and 100 synthetic from Sora, Runway Gen-3, Synthesia, and HeyGen.
The Detection Landscape
Forensic analysis tools examine pixel-level artifacts. Biometric tools focus on facial geometry and micro-expressions. Provenance tools check for C2PA Content Credentials. The most effective strategies combine all three.
Tool Assessments
Sensity AI: 87% accuracy - highest in our evaluation. $500/month. Best for large newsrooms. Reality Defender: 83%. $300/month. Best for fast inline verification. Microsoft Video Authenticator: 81%. Free for newsrooms. Best for Microsoft ecosystem users. Intel FakeCatcher: 79% overall, 91% on face close-ups. Novel photoplethysmography approach. Free academic license. Deepware Scanner: 74%. Open-source, runs locally. Best for sensitive footage. Content Authenticity Verify: Not a detector per se - verifies C2PA provenance. Strongest signal when credentials present. Free.
What Tools Miss
No tool reliably detected AI audio overlaid on authentic video - voice cloning remains a blind spot. Full-scene generation from Sora posed the biggest challenge. Compressed video degrades all tools' accuracy by 10-20 points.
Building a Workflow
Check for C2PA provenance first. Run forensic detection with at least two tools. Apply traditional verification. For high-stakes content, consult a human expert. The cost of publishing a deepfake is reputational damage that can take years to repair.