Distributor LANDR is extending the machine-learning system it built with Montreal's Mila institute to catch streaming fraud, and says the artificial streams Spotify flagged on music it distributes fell by 95% between early 2025 and the second quarter of 2026. It credits several anti-fraud measures together for that drop, not the model alone.
The system has two parts, one scoring accounts on profile and behaviour, the other looking for listening that doesn't resemble a real audience. It will now also look for warning signs before release: very frequent uploads, whole catalogues appearing overnight, linked accounts sharing devices or payout details, and metadata built to pass as a known artist. People review anything flagged before an account is closed, since a sudden spike can simply be a real hit.
The models sit alongside audio fingerprinting from ACRCloud and Ircam Amplify's checks for altered or synthetically generated recordings, plus new identity checks for heavy uploaders. Artists declare AI-generated content when they submit, and that label travels to streaming services in the release metadata.
Fraud costs distributors directly, since Spotify has charged labels and distributors per track for blatant artificial streaming since 2024. LANDR, which serves more than 8 million musicians and bought Reason Studios this year, says the work follows IFPI's new Streaming Integrity Initiative, which MBW understands it may formally join later.
Sources · 1