Skip to content

AI_PROTECTION_GUIDE.md

Latest commit

 

History

History
354 lines (277 loc) · 9.46 KB

AI_PROTECTION_GUIDE.md

File metadata and controls

354 lines (277 loc) · 9.46 KB
Raw

AI Training Protection & Plagiarism Detection System

ProofStamp Enterprise v2.1 - PhD-Level IP Protection for Digital Creators in India

Overview

ProofStamp now includes comprehensive AI training protection and advanced plagiarism detection, making it the most sophisticated copyright protection platform in India.

Core Features

1. AI Access Control Token System

  • Generate unique access tokens for each stamped work
  • Define granular licensing restrictions
  • Prevent unauthorized AI training
  • Track all AI platform access attempts
  • Revoke access at any time
POST /api/ai-protection/access-token
{
  "stampId": "PS-2026-XXXXX",
  "licenseType": "all-rights-reserved",
  "restrictions": {
    "allowAI": false,
    "allowCommercial": false,
    "allowModification": false,
    "attributionRequired": true
  }
}

2. AI Registry Monitoring

  • Monitor Hugging Face datasets and models
  • Track Replicate model registry
  • Monitor CivitAI community models
  • Scan Stability AI integrations
  • Track Kaggle datasets
  • Search GitHub repositories

Automatically scan every 6 hours for your protected works appearing in AI training.

POST /api/ai-protection/registry-monitor
{
  "stampId": "PS-2026-XXXXX",
  "platforms": ["hugging_face", "replicate", "civitai", "kaggle"],
  "scanFrequency": "weekly"
}

3. AI Training Detection

  • Detect if your work is in LAION-5B, OpenImages, COCO, ImageNet
  • Embedding similarity analysis against published models
  • Metadata analysis of model cards
  • Legal action support with evidence
GET /api/ai-protection/analysis/:stampId

4. Deepfake & Manipulation Detection

  • Face consistency analysis
  • Deepfake likelihood scoring
  • AI-generated image detection
  • Compression artifact analysis
  • Comparison with original
  • Temporal video analysis
POST /api/ai-protection/deepfake-analysis
{
  "stampId": "PS-2026-XXXXX",
  "file": <binary>,
  "fileType": "image|video"
}

5. Plagiarism Tracking Across Web

  • Deep web scanning
  • Blockchain explorer monitoring (NFT theft)
  • Cross-platform similarity matching
  • AI-generated image detection
  • Manipulation detection

API Endpoints

Access Control Tokens

# Generate new access token
POST /api/ai-protection/access-token

# List all tokens for a stamp
GET /api/ai-protection/access-tokens/:stampId

# Verify token (PUBLIC - no auth)
POST /api/ai-protection/verify-token

# Revoke token
DELETE /api/ai-protection/access-token/:tokenId

Registry Monitoring

# Enable monitoring for a stamp
POST /api/ai-protection/registry-monitor

# Scan AI registries now
POST /api/ai-protection/scan/:stampId

# Get detected instances
GET /api/ai-protection/detections/:stampId

# Report unauthorized use to platform
POST /api/ai-protection/report-detection/:detectionId

# Disable monitoring
DELETE /api/ai-protection/registry-monitor/:stampId

Analysis & History

# Analyze dataset usage
GET /api/ai-protection/analysis/:stampId

# Get complete detection history
GET /api/ai-protection/history/:stampId

Authentication

All endpoints except /verify-token require authentication:

  • JWT Bearer token: Authorization: Bearer <token>
  • API Key: X-ProofStamp-Api-Key: <key> or X-Api-Key: <key>

Legal Compliance (India)

Basis in Indian Law

  1. Information Technology Act, 2000

    • Section 65: Tampering with computer source documents
    • Section 67: Punishment for publishing obscene material in electronic form
    • Section 72: Reasonable security practices and procedures

  2. Bharatiya Nyaya Sanhita (BNS) 2023

    • Section 63: Electronic Records in evidence
    • Our AI protection acts as a tamper-proof evidence system

  3. Copyright Act, 1957

    • Section 13: Infringement of copyright
    • Our platform documents original authorship and ownership

  4. Information Technology Rules, 2021

    • Rule 3(1)(b): Intermediary due diligence
    • Our takedown automation complies with Rule 3(1)(b)(iv) notice format

Evidence for Court

Our AI protection system provides:

  • Timestamp Authority Proof (RFC 3161): Court-admissible proof of creation
  • Blockchain Anchor: Immutable ledger entry
  • AI Access Token: Licensing proof
  • Detection Records: Detailed evidence of unauthorized training
  • Audit Trail: Complete chain of custody
  • Creator Attestation: Sworn statement of originality

Database Schema

New Tables

AIAccessToken: Licensing control tokens

- id: String (PK)
- stampId: String (FK to Stamp)
- tokenHash: String (unique)
- tokenPrefix: String (unique, shown to user)
- licenseType: String (all-rights-reserved | cc-by | cc-by-sa | ...)
- restrictions: JSON (allowAI, allowCommercial, allowModification, ...)
- verificationCount: Int (tracks AI platform access attempts)
- lastVerifiedAt: DateTime
- revokedAt: DateTime (null = active)

AIRegistryMonitor: Monitoring configuration

- stampId: String (FK to Stamp)
- monitoringPlatforms: JSON (enabled platforms and status)
- scanFrequency: String (hourly | daily | weekly | monthly)
- nextScanAt: DateTime (when next scan runs)
- isActive: Boolean

AITrainingDetection: Detected unauthorized training

- stampId: String (FK to Stamp)
- platform: String (hugging_face | replicate | civitai | ...)
- modelId: String (model identifier on platform)
- modelName: String
- modelUrl: String
- detectionMethod: String (hash_match | embedding_similarity | metadata_analysis)
- confidence: Float (0-1)
- responseStatus: String (new | investigating | contacted | resolved)
- reportedAt: DateTime

TrainingDataAudit: Detailed usage evidence

- stampId: String (FK to Stamp)
- auditType: String (usage_in_dataset | model_training | fine_tuning)
- platform: String
- evidence: JSON (raw evidence data)
- severity: String (low | medium | high | critical)
- actionTaken: String

Background Jobs

AI Registry Scan Job

  • Frequency: Every 6 hours
  • Function: Scan all enabled AI monitoring stamps
  • Detection: Searches for stamps in datasets/models
  • Notification: Alerts creators when unauthorized usage detected
  • Action: Automatically creates detections for legal action

Configure via environment:

AI_REGISTRY_SCAN_INTERVAL=6h  # hours between scans
AI_REGISTRY_BATCH_SIZE=50     # stamps scanned per run

Microservices

Deepfake Detection Service (Python)

Runs on port 5000 as separate Flask microservice:

python stego-service/deepfake_detection.py

Endpoints:

  • /analyze/face-consistency - Face anomaly detection
  • /analyze/deepfake-score - Deepfake likelihood
  • /analyze/ai-generated - AI generation detection
  • /analyze/artifacts - Compression/forensic analysis
  • /analyze/compare - Comparison with original
  • /video/extract-frames - Video frame extraction

Environment:

DEEPFAKE_DETECTION_SERVICE_URL=http://localhost:5000

Implementation Roadmap

Phase 1: Implemented ✅

  • ✅ AI Access Token generation and verification
  • ✅ AI Registry monitoring (Hugging Face, Replicate, CivitAI, Kaggle)
  • ✅ Training detection infrastructure
  • ✅ Deepfake detection framework
  • ✅ Audit and notification systems

Phase 2: In Development 🔄

  • 🔄 Complete deepfake detection ML models
  • 🔄 Audio/video fingerprinting
  • 🔄 C2PA full embed+verify workflow
  • 🔄 Fair use detection filtering

Phase 3: Coming Soon 📅

  • 📅 Dark web monitoring
  • 📅 Blockchain marketplace scanning
  • 📅 Auto-takedown for all platforms
  • 📅 Litigation dashboard
  • 📅 Advanced analytics

Best Practices

For Creators

  1. Enable AI Opt-Out Registry

    • Create stamp
    • Go to AI Protection tab
    • Enable "Prevent AI Training" (default ON)
    • Generates auto-public entry in /registry

  2. Set Up Monitoring

    • Enable registry monitoring (weekly)
    • Receive notifications if work detected
    • Review detections in dashboard
    • Report to platforms with one click

  3. License Your Work

    • Choose appropriate license type
    • Generate access token if allowing some AI use
    • Share restrictions with platforms
    • Revoke token if policy changes

  4. Track Evidence

    • Export detection history regularly
    • Store in safe location
    • Use for legal proceedings
    • Share with lawyers

For Platform Integration (AI Model Creators)

  1. Check License Before Training

    import requests

# Verify if content can be used for training
response = requests.post('https://api.proofstamp.app/ai-protection/verify-token', 
  json={'token': user_token})

if not response.json()['allowTraining']:
  # Don't use in training
  pass

  2. Respect Opt-Out Registry

    # Check if work is protected
response = requests.get('https://api.proofstamp.app/registry/check',
  params={'hash': file_sha256, 'phash': perceptual_hash})

if response.json()['protected']:
  # Exclude from training
  pass

  3. Implement robots.txt directives

    User-Agent: anthropic-ai
Disallow-Training: /

Support & Legal

For legal questions about AI training protection in India:

Compliance Verification

All timestamps verified against Indian Government standards:

  • TSA Provider: Digital Signature algorithms per DISHA standards
  • Blockchain: OpenTimestamps (ISO 18014-3)
  • Jurisdiction: India (applicable throughout)

ProofStamp - Trust Built on Law, Technology, and Time.