LANDR says artificial streams flagged by Spotify across the music it distributes fell 95% between Q1 2025 and Q2 2026, as it expands a fraud detection system built with Montreal's Mila – Quebec AI Institute. The company credits the drop to several fraud-prevention measures, not the Mila models alone, and didn't disclose how many streams were flagged before or after.
The system pairs two machine-learning models, one scoring accounts on profile and behavioral data and another scanning listening activity for patterns that don't look like a real audience.
The expansion pushes detection earlier, before releases reach streaming services. New risk signals include unusually high release frequency, catalogs that appear overnight, clusters of accounts sharing devices or payout details, and metadata built to impersonate established artists. LANDR says its anti-fraud team reviews every flag before acting, and it's adding identity verification for accounts that release frequently. Artists already declare AI-generated content at upload, and LANDR says that disclosure travels in the metadata sent to streaming services.
The timing lines up with a broader push to make distributors the choke point. IFPI launched its Streaming Integrity Initiative on Sept. 14, asking distributors to verify customers, vet uploads, act against repeat offenders, share intelligence and measure results. LANDR says its expansion advances the same priorities but isn't among the launch signatories. Neither is DistroKid, which Universal sued this month over the AI tracks it passes to streaming services.
The fraud and AI stories keep converging. Deezer says up to 85% of streams on fully AI-generated tracks were fraudulent in 2025, which is why distributors are being asked to police the pipeline before the plays happen.
For AI music creators, the signals matter more than the headline number. High release frequency and fast-growing catalogs describe a lot of legitimate AI artists, not just fraud farms. LANDR's promise of human review before enforcement, and CEO Pascal Pilon's point that a sudden spike can simply be an artist having a good day, will be tested hardest on exactly those users.