LANDR says artificial streams flagged by Spotify across music it distributes fell 95% between the first quarter of 2025 and the second quarter of 2026, as the company adds more fraud screening before releases reach streaming services.
The Montreal-based music-creation and distribution platform disclosed the figure in a September 22nd update to technology developed with Mila, the Quebec AI Institute. LANDR says its newer system can identify potentially suspicious accounts and releases earlier, giving its anti-fraud team an opportunity to investigate before music is delivered to Spotify and other digital platforms.
The 95% figure comes with an important qualification. LANDR attributes the decline to multiple fraud-prevention measures introduced during the period rather than solely to its latest machine-learning models. The Mila-developed system was deployed in February 2026, more than a year after the measurement period began, LANDR says.
What the number does show is that Spotify identified substantially less artificial activity across LANDR-distributed music by the end of the measured period. The distributor is now trying to move more of that detection upstream, before suspicious releases have an opportunity to generate artificial streams at all.
The 95% Figure Measures Spotify Flags
LANDR is not claiming that 95% of all streaming fraud involving its customers disappeared, nor is the figure an independent audit of every play generated by music distributed through the company.
The measurement is narrower: artificial streams subsequently flagged by Spotify across LANDR-distributed music declined by 95% between Q1 2025 and Q2 2026.
That distinction is important because Spotify is making the downstream determination about which listening activity it considers artificial. A large reduction in those flags indicates that substantially less LANDR-distributed activity reached the point where Spotify identified it as manipulated.
Music Business Worldwide highlights the same measurement period and notes that LANDR introduced several anti-fraud initiatives across it. The Mila models were only deployed in February, making it impossible to attribute the entire 95% decline to that technology alone.
The scale of LANDR’s operation helps explain why automation is involved. The company says its platform serves more than 8 million musicians, making manual screening of every account and release impractical.
LANDR chief technology officer Tommy Desilets described the challenge as a “needle-in-a-haystack problem.” Machine learning is intended to narrow that haystack before investigators decide what actually requires action.
Two Models Look for Different Warning Signs
The core system came out of a six-month collaboration between LANDR and Mila, the Montreal research institute founded by deep-learning researcher Yoshua Bengio.
LANDR says the project produced two complementary machine-learning models. One evaluates account-level information and behavioral patterns. The other examines streaming activity for signals that do not resemble a genuine audience.
The company has since expanded the information available to its fraud operation. Warning signs can include unusually high release frequency, entire catalogs appearing almost immediately, groups of accounts sharing devices or payout details and metadata that appears designed to impersonate an established artist.
LANDR has also introduced identity-verification measures that can require users to confirm who they are during signup or later when an account triggers additional review. Its verification documentation connects those checks with efforts to prevent streaming fraud, copyright infringement, impersonation and bot-driven royalty manipulation.
None of those signals automatically proves fraud.
A label legitimately moving a catalog could deliver many releases in a short period. An artist can suddenly gain attention because a song takes off on social media or receives unexpected exposure. Shared account characteristics may also require context before they indicate anything improper.
That is why LANDR places human review between automated detection and enforcement. “Sometimes a song suddenly taking off is exactly that: an artist having a really good day,” CEO Pascal Pilon said in the company announcement.
Accounts confirmed to be fraudulent can be closed and prevented from returning. LANDR says the results of those investigations can then be used to improve future detection.
Spotify Already Penalizes Artificial Streaming
LANDR has a financial reason to catch severe manipulation before Spotify does.
Since April 1st, 2024, Spotify has charged labels and distributors when it detects flagrant artificial streaming on tracks they deliver. The company says the fee applies only when artificial activity reaches very high levels on a recording and is intended to give distributors an incentive to identify repeat offenders before their music reaches the service.
Spotify defines artificial streaming as listening that does not reflect genuine user intent, including attempts to manipulate streams through automated systems such as bots or scripts. Its artificial-streaming guidance says confirmed artificial streams do not earn royalties and do not provide the same benefits in public stream counts, charts or recommendation systems.
Serious cases can also lead to playlist removal, action from a distributor or removal of a track from Spotify.
The policy creates a direct incentive for companies sitting between artists and DSPs to improve their own screening. Waiting until Spotify identifies flagrant manipulation means the suspicious activity has already happened and can potentially create penalties for the distributor.
LANDR’s newer approach attempts to move part of that work to the upload and distribution stage instead.
Fraud Detection Moves Toward Distribution
Digital distribution dramatically lowered the barrier between independent musicians and global streaming services. An artist no longer needs a traditional record deal to deliver music to Spotify, Apple Music and other major platforms.
The same openness can make distributors an attractive entry point for fraud operations.
At the extreme end, bad actors can create or acquire recordings, distribute them through legitimate services and then use bots, automated accounts or streaming farms to manufacture listening. If the plays qualify for royalties before being detected, the operation can attempt to turn manipulated activity into money.
That is why the industry’s response is increasingly moving toward the companies that deliver music in the first place. On September 14th, IFPI launched the Streaming Integrity Initiative, backed by major label groups, independent companies and distributors.
The initiative organizes its baseline anti-fraud practices around five areas: Verify, Vet, Act, Share and Measure. Identity checks, content screening, enforcement, information sharing and measurement are all intended to make manipulation harder at multiple points in the supply chain.
LANDR’s expanded system follows that same upstream logic. Detect a suspicious account or release before delivery and Spotify never has to identify the resulting fake streams afterward.
For a distributor serving millions of creators, machine learning makes that kind of screening possible at a scale a human team alone could not realistically match. Human judgment then becomes most important where the system finds something unusual.
Artificial Streams Can Redirect Real Money
Fake streaming can make a recording appear more popular than it really is, but inflated numbers are only the visible part of the problem.
Streaming fraud can also become royalty fraud. Automated listening can attempt to generate revenue for recordings that are not attracting genuine listeners, potentially diverting money within the streaming economy toward manipulated activity.
Spotify says streams it identifies as artificial do not earn royalties. Catching manipulation therefore determines whether suspicious listening produces a financial benefit rather than merely whether a public number gets corrected.
The scale problem is becoming more important as fraud operations gain access to tools that make large catalogs easier and cheaper to create. IFPI’s 2026 industry report warns that generative AI can make high-volume fraud easier by reducing the cost of producing content that can then be paired with automated listening.
That does not make AI-generated music fraudulent by definition.
A fraud operation can use entirely human-created recordings, and an AI-assisted track can attract a completely legitimate audience. The relevant question for artificial streaming is whether the listening reflects real user intent.
LANDR itself permits music created with AI assistance under its distribution rules while applying restrictions intended to address intellectual-property, impersonation and high-volume spam risks.
The company also uses audio-recognition technology from ACRCloud and IRCAM to identify separate problems such as commercial recordings being re-uploaded, altered copies and certain signals associated with synthetic content. Those systems address what is being uploaded; LANDR’s fraud models also examine who is uploading it and the behavioral patterns surrounding the account.
Legitimate Artists Can Still Be Flagged
Artificial streams appearing on an artist’s music are not proof that the artist deliberately committed streaming fraud.
Independent musicians are frequently marketed playlist and promotional services promising exposure. Some openly guarantee streams, while others appear legitimate but may work with partners or playlists that use artificial activity behind the scenes.
LANDR warns that legitimate tracks can even appear on fraudulent playlists without an artist’s permission. Bad actors may mix real music into manipulated playlists to make the listening activity appear more credible.
Its StreamScan product exists partly to give artists visibility into that problem. StreamScan displays artificial-streaming reports LANDR receives from DSPs including Spotify and Apple Music so musicians can see when their released tracks have been flagged.
StreamScan does not itself decide whether those streams are artificial. It tells artists what DSPs have already reported. LANDR’s newer fraud-detection system operates earlier, looking for suspicious accounts and releases before or during distribution.
That distinction matters because false positives carry real consequences. A detection system optimized only to reduce Spotify flags could become too aggressive if unusual release patterns or sudden legitimate success were automatically treated as wrongdoing.
LANDR says human investigators remain responsible for reviewing suspicious cases rather than allowing the machine-learning models to make every enforcement decision themselves.
For independent musicians, that human layer may become increasingly important as distributors add more automated screening between an upload and its arrival on streaming services.
The Upload Becomes the Next Fraud Checkpoint
LANDR’s 95% figure does not show that streaming fraud has disappeared. It is a company-reported measurement covering Spotify-flagged artificial streams across LANDR-distributed music during a specific period, and several fraud-prevention measures changed along the way.
The result does show why distributors are becoming a larger part of the anti-fraud infrastructure.
Spotify can remove the financial and promotional benefits of artificial streams once it identifies them. LANDR’s strategy is increasingly about reducing the number of suspicious releases and accounts that ever reach that stage.
Identity verification adds friction before distribution. Machine-learning models narrow millions of accounts and releases into cases worth investigating. Human reviewers decide whether unusual behavior actually warrants enforcement. StreamScan then gives legitimate artists more visibility when DSPs flag artificial activity after release.
Those pieces attack the problem at different points rather than relying on one detection model to solve it.
LANDR’s reported 95% decline therefore should not be read as evidence that its new AI system eliminated streaming fraud. It shows that Spotify flagged substantially less artificial activity across music distributed by LANDR between Q1 2025 and Q2 2026 while the company added multiple layers of prevention.
The next stage of that strategy happens even earlier: identifying risky accounts, suspicious catalogs and unusual behavior before the music reaches Spotify at all.