Humark's integrity does not rely solely on the honesty of registrants. The platform includes passive forensic systems that detect anomalous registration patterns.
Perceptual Hash (pHash) Analysis
When an asset is registered, Humark generates both a cryptographic hash (SHA-256) and a perceptual fingerprint (pHash). The perceptual hash captures the visual structure of the image in a way that survives minor modifications like compression or resizing. At scale, a registry of millions of pHash values creates a statistical corpus that reveals patterns invisible at the individual level.
AI-Generated Image Signatures
AI-generated images have characteristic frequency distributions at the DCT (Discrete Cosine Transform) level. When the adversarial shield processes an image, it captures the spectral signature of the shielded asset. These signatures differ systematically between organic human-created art and AI-generated outputs.
Temporal Clustering Detection
Mass registrations from a single account, similar perceptual hashes registered in rapid succession, and batch pipeline behavior patterns are automatically flagged for review. A genuine artist registers work sporadically over time. A fraudulent actor running a generation pipeline produces telltale temporal patterns.
AI Provenance Signal (v2 Roadmap)
Future versions of Humark will include a passive AI-origin probability classifier at registration time. This is not a gate. It is a risk signal appended to the manifest (e.g., humark.ai_probability: 0.91). The registry is honest about uncertainty. This is better than binary gating because it is falsifiable and transparent.
These forensic layers work together to create a living, improving system, not a static gate that can be bypassed once and forgotten.