First Streaming Fraud Case Nears Sentencing, Global Publishing Revenues Hit $12.2B, and Appeals Court Rejects AI Fair Use Defense

How The Music Business Works - Issue #81

October 1, 2026

Welcome back to How the Music Business Works.

A federal appeals court ruled this week that using copyrighted material to train AI models does not qualify as fair use, a precedent-setting decision that strengthens the music industry's hand in its ongoing litigation against Suno and Udio. On the same front, IMPF and IMPEL published a framework demanding that songs be valued at least equally with recordings in AI licensing deals, pushing back against the imbalances that have long favored labels over publishers. 

Global music publishing revenues hit $12.2 billion in 2025, outpacing recorded music growth and signaling a sector with strong momentum heading into the AI era. 

In Japan, a Tokyo court ruled for the first time that a person's voice is a protected publicity right, even as it declined to order the removal of the AI-cloned content at issue. And in the US federal prosecutors are seeking at least 46 months in prison for Michael Smith, the architect of the first criminal streaming fraud case in US history.

Here’s the news.

Prosecutors Seek 4 Year Prison Sentence in First US Streaming Fraud Criminal Case

Federal prosecutors are asking a judge to sentence Michael Smith to at least 46 months in prison for a streaming fraud scheme in which bots streamed AI-generated songs billions of times, generating more than $8 million in royalties he was not entitled to. Smith, who pleaded guilty in March, is due to be sentenced on October 6 in Manhattan federal court. Prosecutors are also seeking forfeiture of the full $8.09 million in fraudulent proceeds. The scheme, which ran for more than six years starting in 2017, involved up to 10,000 bot accounts registered under fictitious names, bulk-purchased email addresses, and corporate debit cards obtained through false pretenses. To avoid detection, Smith spread artificial streams across hundreds of thousands of songs and used AI to generate enough content to keep individual stream counts low. Prosecutors note that in April 2023 alone, Smith's catalog received nearly nine times more family-plan streams on YouTube Music than Taylor Swift's entire catalog. Smith's lawyers are asking for probation, arguing that no single victim lost a significant amount, but prosecutors counter that every penny of the loss went directly into Smith's pocket. As the first criminal prosecution for streaming fraud in the US, the sentence is expected to set a significant precedent for the industry.

Global Music Publishing Revenues Hit $12.2 Billion in 2025

Global music publishing revenues grew 9.6% to $12.2 billion in 2025, outpacing the 6.4% growth recorded by the wider recorded music market, according to Midia Research. Major publishers regained ground over independents, growing their combined share from 53.4% to 54.7%, with Sony Music Publishing accounting for nearly a full percentage point of that shift. Music catalog funds also had a strong year, with revenues up 14.9% to $1.6 billion. Streaming remains the largest source of publishing revenue and is expected to grow its share further through 2033. Midia also sees AI as an opportunity for publishers to secure a larger slice of revenues relative to recordings than they have in streaming, where the imbalance has long favored labels. The broader message from the report is that the publishing sector's next phase will reward active catalog management rather than passive rights accumulation, with analysts pointing to AI, global expansion, and new use cases as the key battlegrounds ahead.

Japanese Court Rules Voice Is a Protected Publicity Right in First AI Cloning Case

A Tokyo court has ruled for the first time in Japan that a person's voice is a protected publicity right, in a case brought by voice actor Kenjiro Tsuda against TikTok over nearly 200 videos narrated by an AI voice he claimed was based on his own. Tsuda, known for his roles in Jujutsu Kaisen and Yu-Gi-Oh!, argued the AI voice was deliberately similar to his and misled viewers, while the anonymous uploader was allegedly earning up to $4,800 a month from the content. Despite the landmark ruling on voice rights, the court dismissed Tsuda's request to have the videos removed, in part because the account had already been deleted. TikTok maintained the narration used a generic male voice that would not be mistaken for Tsuda. The case marks the first time a Japanese voice actor has taken a major company to court over alleged AI voice cloning, and arrives as Japan's Justice Ministry has begun laying out guidelines on voice rights in the context of AI-generated content.

Appeals Court Rules AI Training on Copyrighted Works Is Not Fair Use

A US federal appeals court has upheld a ruling that using copyrighted material to build AI training data does not qualify as fair use, in a decision marked as precedential by the Third Circuit. The court sided with Thomson Reuters in its case against Ross Intelligence, a now-defunct startup that copied Westlaw's editorial summaries to train a competing AI-powered legal search tool. The RIAA and NMPA had filed a joint brief in the case, arguing that training an AI model on copyrighted works to create a competing product can never be fair use, and pointing to the growing market for licensed AI training data as evidence that alternatives exist. The 32-page opinion has been filed under seal pending proposed redactions, but once unsealed it will bind district courts across the Third Circuit. While the ruling does not directly address generative AI, it establishes key appellate precedent at a moment when the fair use question is at the center of multiple ongoing disputes between the music industry and AI companies.

IMPF and IMPEL Push for Equal Song and Recording Rights in AI Licensing

Independent music publishing organizations IMPF and IMPEL have published a framework for licensing generative AI, with one non-negotiable principle at its core: songs must be valued at least equally with recordings. The demand is a direct challenge to the streaming era's longstanding imbalance, in which recording rights have consistently received a significantly larger share of revenue than publishing rights. With AI deals now being struck at a regular pace, the organizations argue this is the moment to establish a fairer framework rather than carry old assumptions forward. The framework also calls for full transparency throughout the licensing process, including clear distinctions between payments for past use, training, outputs, and future exploitation, and fair treatment of any deductions or revenue calculations. IMPF and IMPEL acknowledge the complexity of the landscape, where deals may involve multiple income streams covering training data, AI-generated outputs, and the separate exploitation of artists' voices. The organizations say they are ready to engage constructively with AI companies and technology providers, but insist that emerging business models must not repeat the structural imbalances that have long disadvantaged songwriters in the digital market.

Random Ramblings:

• Three Ways To NOT Get Paid for Your Music.
• Why Stream Counts Don’t Tell the Full Story of Music Marketing Performance.
• Taylor Swift dominated the VMAs, but do music videos still matter?
• What It Actually Takes to Read Music Without Sight.