DGR embeds collusion-resistant forensic watermarks directly into AI model weights, PDFs, and documents. When a file leaks, we attribute it to the exact partner — even if multiple partners colluded.
Differential spread-spectrum watermarking, canary tokens that phone home on open, real-time threat scoring, tamper-evident evidence chains, automated DMCA, and a live forensic lab — all in one platform.
Each layer solves a different part of the leak problem. Together they produce legally-admissible attribution evidence.
Every partner download gets a mathematically unique 128-bit LSB signature across 131,072 float32 positions. Spread-spectrum coded — resists averaging by up to ~32 colluders.
All plansEvery file carries an invisible phone-home token. The moment a partner opens it — anywhere — you get an alert with their IP and timestamp. Before any re-download attempt.
All plansEvery download scored in real time: Tor (+60), impossible travel (+45), datacenter (+20), automation (+40). High-risk downloads auto-corrupted. Critical scores hard-blocked.
All plansEvery event chained via HMAC-SHA256. Retroactive tampering is detectable. Export as a signed JSON package for legal proceedings.
AI Lab+Deploy decoy high-value files that look legitimate. When accessed, fire an instant alert identifying insider threats and external adversaries before real files are touched.
Team+Monitor HuggingFace, GitHub, and Telegram for leaked files. One-click DMCA generation, automated takedown filing, and status tracking. Integrated with forensic evidence export.
AI Lab+Per-file country allowlists and blocklists. Compliance presets (GDPR, OFAC). Real-time country detection on every download. Override per-partner for exceptions.
Team+Upload a suspected leaked file and extract embedded watermarks in seconds. Get attribution with confidence scores, evidence report, and pre-filled legal documents.
AI Lab+Upload any file — AI model checkpoints, PDFs, code, datasets. Set download limits, expiration windows, geo-fences, device allowlists, and approval workflows. AES-256-GCM encryption at rest.
Generate a unique access link per partner. Each partner receives their own differential watermark signature, canary token, and forensic envelope. Links are email-verified with OTP.
Before serving any file, DGR scores the request against Tor exit lists, geo-velocity, datacenter IP ranges, and download frequency. Suspicious requests receive corrupted files or are blocked.
When a partner opens the file anywhere, the embedded canary token pings DGR. You see the IP, timestamp, and user agent — even before any re-download attempt.
Submit a leak URL or upload the suspected file to the Forensic Lab. DGR extracts the differential watermark, correlates against partner signatures, and returns an attribution with confidence score.
One click generates a signed evidence package, pre-fills a DMCA notice, and files the takedown request automatically. Track status in real time. Export for legal proceedings.
Every download attempt is scored before any file is served. No client software required.
Six independent risk factors evaluated in under 200ms per request using live IP intelligence.
Scores map to automatic actions — no manual review required for standard cases.
THREAT_BLOCKED code, logged to evidence chainAll plans include differential watermarking, canary tokens, and threat scoring. Higher tiers unlock forensic analytics, geo-fencing, honeypots, and automated DMCA workflows.
DGR is designed for attribution and deterrence, not absolute prevention. No server-side system can stop a determined adversary from photographing their screen, using a hardware capture device, or exfiltrating via a compromised endpoint after authorized access is granted. DGR maximizes the probability that leaks are attributed (forensic watermark + canary correlation), increases the legal strength of that attribution (HMAC-signed evidence chain), and deters casual unauthorized redistribution (silent corruption, threat scoring, geo-fencing, honeypots). It is an attribution-first distribution engine — not a DRM system.