50 lines (40 loc) · 2.36 KB

GDPR / Right-to-Erasure Compliance: AI Abstraction Review (#301)

Scope

Review of the AI feature set (artist AI tools, recommendation engine, analytics) for GDPR Article 17 (Right to Erasure) compliance.

Current AI Data Processing

1. AI Artist Tools (Per-Artist Opt-in/Opt-out)

Data collected: Song metadata, listening patterns, genre tags
Storage: PostgreSQL (ai_preferences table per artist)
Erasure: ✅ Artist can toggle opt-out; data deleted on account deletion
Recommendation: Ensure AI model retraining excludes erased user data

2. Recommendation Engine

Data collected: User listening history, saves, playlist additions
Storage: Redis cache + PostgreSQL (user_interactions table)
Erasure: ⚠️ Redis cache has TTL but PostgreSQL records need explicit deletion
Action needed: Add cascade delete from user_interactions on user deletion

3. Analytics & Metrics

Data collected: Play counts, geographic distribution, engagement scores
Storage: PostgreSQL (song_analytics table), aggregated
Erasure: ⚠️ Aggregated metrics cannot be un-aggregated; document this limitation
Action needed: Add note that aggregated analytics survive erasure per Art. 17(3)(b)

Erasure Checklist

Data Store	Erasure Method	Status
`ai_preferences`	CASCADE DELETE on user	✅ Implemented
`user_interactions`	Explicit DELETE on user deletion	⚠️ Needs implementation
Redis recommendation cache	TTL-based expiry	✅ Working
`song_analytics` (aggregated)	Cannot erase (Art. 17(3)(b))	✅ Documented
Model training data	Retraining exclusion list	⚠️ Needs implementation
AI-generated content	Keep (Art. 17(3)(d) - public interest)	✅ Documented

Recommendations

Add user_erased flag to prevent AI from processing erased users' data
Update recommendation query to exclude user_erased = true users
Document AI data retention in privacy policy
Implement audit log for AI data access (who queried what, when)

Acceptance Criteria

AI opt-out toggle works for artists
User deletion cascades to AI preferences
User deletion cascades to user_interactions (action needed)
Recommendation engine excludes erased users (action needed)
Privacy policy documents AI data retention