The AI Music Debacle: Lawsuits, Deepfakes, and the Fight Over Who Owns a Voice

If 2023 was the spark and 2024 was the lawsuits, then 2025 is the year the whole thing caught fire at once. AI music stopped being a story about a clever fake Drake song and became a genuine crisis touching copyright law, streaming economics, artists' livelihoods, and the basic question of whether you can trust that the band you love is made of people.

Here's an honest look at the debacle — what actually went wrong this year, and why almost nobody is happy.

The short version

• A Copyright Office report gutted the AI companies' main legal defence.
• A fake "band" fooled a million Spotify listeners — with no label to warn them.
• Streaming got flooded: on some platforms, AI tracks make up nearly a fifth of daily uploads, most of their streams fraudulent.
• Spotify finally cracked down — and pulled tens of millions of tracks.
• Europe joined the legal fight, and the lawsuits got existential.

The legal ground shifted

For two years, every AI music company leaned on the same defence: training on copyrighted songs is fair use — transformative, harmless, basically the digital version of a musician learning by listening.

In May 2025, the US Copyright Office took that argument apart in a detailed report on AI training. It rejected the claim that ingesting copyrighted work is "inherently transformative," and zeroed in on the part that matters most to working musicians: market harm. If a model trains on your catalogue and then floods the market with cheap substitutes that compete directly with you, that's not abstract — that's your income.

It's not a court ruling, and it doesn't end the lawsuits. But it badly weakened the AI side's footing right as the cases against Suno and Udio ground toward their decisive phase. And the fight stopped being only American: rights organisations in Europe — Germany's GEMA, Denmark's Koda — opened their own actions, arguing the same thing in different jurisdictions. This is now a global legal pile-up.

The Velvet Sundown: a million people, fooled

The most quietly unsettling moment of the year wasn't a lawsuit. It was a band that didn't exist.

In mid-2025, a psych-rock group called The Velvet Sundown started climbing Spotify fast — past a million monthly listeners. Slick songs, a backstory, album art, the works. The only problem: there were no musicians. The entire project was AI-generated, and for weeks nobody listening had any way to know.

When Deezer ran it through its detection tool, the verdict was 100% AI. The scandal wasn't really that someone made an AI band — it was that the platform served it to a million people with no label, no disclosure, nothing. If you can't tell whether a song was made by a person, the basic contract between listener and artist quietly breaks. The Velvet Sundown became the case study every disclosure debate now points to.

The flood

Behind one viral fake band sits an ocean of anonymous ones.

By late 2025, Deezer — one of the few platforms publishing real numbers — reported that around 18% of the tracks uploaded to it every day were fully AI-generated: more than 20,000 songs daily, a figure that had tripled in two years. Worse, Deezer estimated that up to 70% of the streams on those AI tracks were fraudulent — bots, not people, gaming the royalty system in exactly the way that first criminal fraud case foreshadowed.

Think about what that does to a streaming pool. Royalties are split from a finite pot. Every fraudulent stream on a machine-made track is a fraction of a cent siphoned away from a human artist who actually has rent to pay. The threat was never just aesthetic. It's mathematical.

Spotify finally moves

For most of this, Spotify — the platform that matters most — stayed conspicuously quiet, declining to label or detect AI tracks even as Deezer did both.

In September 2025, it finally acted, announcing three measures:

1. An impersonation crackdown. Unauthorised AI voice clones and vocal deepfakes are banned; using a recognisable copy of another artist's voice now requires their documented consent.
2. A spam filter targeting the bad actors mass-uploading near-duplicate, SEO-stuffed, bot-farmed tracks to game recommendations.
3. AI disclosure through a new industry credits standard (developed by the DDEX consortium), so tracks can declare whether AI was used in the vocals, instruments or production.

Tellingly, Spotify also revealed it had removed more than 75 million spammy tracks in the previous year. That number alone tells you the scale of what's being uploaded — and how much of it isn't anyone's idea of art.

Crucially, Spotify framed this as targeting fraud and spam, not responsible AI use. The disclosure won't automatically bury a track. It's a labelling-and-hygiene play, not a ban — which satisfied almost nobody completely, but at least acknowledged the problem out loud.

The human cost

It's easy to lose the people in all the policy. They've been loud, though.

Back in 2024, more than 200 artists — names like Billie Eilish, Stevie Wonder and Nicki Minaj among them — signed an open letter through the Artist Rights Alliance warning against the "predatory" use of AI to devalue musicians and replace human creativity. Through 2025 that anxiety hardened into something concrete: session players watching demo and library work dry up, songwriters seeing their catalogues turned into training fuel without consent or payment, and emerging artists wondering how you get discovered at all when the upload queue is 20,000 machine-made songs deep, every single day.

Not everyone is a doomer. Plenty of musicians use AI tools openly and well, as instruments rather than replacements. But "I used a tool to make my song" and "a company strip-mined my life's work to compete with me" are very different sentences, and most of the year's fights live in the gap between them.

So where's the resolution?

Here's the uncomfortable answer as of autumn 2025: there isn't one yet.

The lawsuits haven't been decided. The NO FAKES Act hasn't passed. Detection is imperfect and disclosure is voluntary. And — the plot twist almost nobody predicted — the major labels that spent a year suing the AI companies are now reportedly negotiating licensing deals with them. The same executives calling AI music theft in court appear to be working out the price of selling it the keys. Whether that becomes a fair settlement for artists or a backroom carve-up over their heads is the question that will define next year.

That's the debacle in one line: a technology moving faster than the law, the economics, or the ethics can keep up — and a lot of people in the middle hoping someone draws the lines before the flood gets any deeper.

We built Videojam around a stubbornly old-fashioned idea: that the antidote to an infinite feed of machine-made content is human judgement — real people pointing at real music videos worth watching. The more synthetic the firehose gets, the more that matters.

This is part of our AI & Music series. Next, we look at where the dust is finally starting to settle — the state of AI music in 2026.