LANDR expands Mila-built fraud system as Spotify flags fall 95%
Music distributor LANDR says the volume of artificial streams flagged by Spotify on music it handles fell 95% between the first quarter of 2025 and the second quarter of 2026.
LANDR, a company that puts tracks onto streaming services for artists and labels, credited several fraud-prevention efforts for the fall rather than one tool alone, according to Music Business Worldwide.
The company announced on September 22 that it has expanded a fraud-detection system built with Mila, the Montreal-based non-profit AI research institute founded by Yoshua Bengio. The first machine-learning models in the collaboration were developed and deployed by Édouard Demotes-Mainard, and LANDR's anti-fraud team has operated and continued to develop the system since then.
Fake streams are plays generated without genuine listener intent, often through bots, stream farms or compromised accounts. The stakes are cash. LANDR says artificial plays can lead to withheld royalties, removed releases and damage to reputation, and Spotify has charged labels and distributors a per-track fee where it finds flagrant artificial streaming since April 2024.
That penalty structure explains why distributors police their own catalogues. A distributor collects and passes on recordings from thousands of independent artists, so one bad actor uploading manipulated tracks can create clean-up work and costs for everyone else in the pipeline.
LANDR's system uses two models working in parallel. One assesses accounts using profile and behavioural data, such as how an account is set up and how it acts. The other examines listening activity for patterns inconsistent with a genuine audience, such as plays that do not look like normal human listening.
The expansion moves checks earlier in the distribution process, before suspect music can gather streams. LANDR said the additional risk signals include unusually high release frequency, catalogues that appear overnight, clusters of accounts sharing devices or payout details, and metadata designed to impersonate established artists.
Those signals target tricks that are hard for listeners to spot. A sudden flood of albums, repeat payout information across supposedly separate accounts, or a new act borrowing the name and credits of a known artist can all suggest automation rather than a developing fan base.
Flagged activity is reviewed by LANDR's anti-fraud team before any action is taken. Accounts confirmed as fraudulent can be closed and blocked from returning, and LANDR Distribution offers a StreamScan dashboard that lets users monitor their own releases for artificial-streaming activity.
What makes this stand out for ordinary listeners is the human check in the loop. The software surfaces risk, people decide, and that decision determines whether music stays up, whether royalties flow, and whether an artist's name remains clear on the services where strangers find it.
