How We Detect AI-Generated Content
One of the key challenges in content verification is distinguishing between human-created and AI-generated content. At Vericum, we use a multi-layered approach to tackle this problem.
Our Detection Approach
1. EXIF Integrity Analysis
Real photographs captured by cameras contain rich EXIF metadata: camera make and model, lens information, exposure settings, GPS coordinates, and more. AI-generated images typically lack this metadata or contain inconsistent data.
Our engine analyzes EXIF completeness and consistency as a strong signal of authenticity.
2. C2PA Manifest Presence
Content with valid C2PA manifests from trusted certificate authorities provides the strongest signal of human origin. This cryptographic proof is extremely difficult to forge.
3. Statistical Analysis
We examine image properties for patterns commonly associated with AI generation, including unusual frequency distributions and artifacts characteristic of diffusion models and GANs.
Scoring System
Our composite verification score weighs these factors:
- **C2PA**: 40% (strongest signal)
- **AI Detection**: 30% (heuristic analysis)
- **EXIF Metadata**: 20% (completeness and consistency)
- **Uniqueness**: 10% (duplicate detection)
Content scoring above 70% is automatically verified. Scores between 40-70% are flagged for manual review. Below 40% results in rejection.
Looking Forward
Post-MVP, we plan to integrate dedicated AI detection APIs for even more accurate classification. The arms race between AI generation and detection continues, and we're committed to staying ahead.