India's labels want streaming services to pay nothing for AI "slop". Who decides what counts?

India's oldest record label has put a simple rule to the world's biggest streaming services: music made purely by AI should earn nothing. On Saregama's earnings call for the quarter to 31 March 2026, managing director Vikram Mehra said labels in India and globally are working with the three leading global streaming platforms so that no value is assigned to purely AI-generated content when the royalty pool is shared out. MediaNama and Music Ally reported the remarks on 19 May 2026.

He didn't name the three platforms. The goal, in his framing, is that pool money goes only to "genuine IP", not to what he called AI slop.

What exactly are the labels asking for?

Mehra wasn't asking for AI music to be banned. He was asking for it to be excluded from payment. As Music Ally's fuller write-up of the call puts it, he argued that artist-driven music will become "far more valuable" as AI content grows, and that mass-generated slop can never give listeners the connection they have with artists.

Two other details from the call change how the demand reads. First, Mehra said there is "no revenue leakage today", and MediaNama reports that he saw no market-share loss to AI either. So this is a rule being set before the damage, not a response to it. Second, Saregama isn't against AI. Mehra said AI is "creating new licensing opportunities" and that Saregama will engage commercially with AI companies at the right time. The label is also testing generative AI to make music videos for catalogue songs by composers such as Hemant Kumar and S.D. Burman.

That sits alongside a fight over training. In February 2025 the Indian Music Industry body, on behalf of labels including T-Series, Saregama and Sony Music, went to the Delhi High Court over OpenAI's use of their songs to train its models without consent, Rolling Stone India reported. The bid was tied to the news agency ANI's earlier case against OpenAI. Put together, the position is consistent: ask us first, pay us, and don't let unlicensed output take a share of the pool.

Has any service actually done this?

One has, publicly. On 29 June, Tidal said it would not knowingly pay royalties on music it identifies as wholly AI-generated, Music Business Worldwide reported. Those tracks stay on the service with an "AI" tag. Tidal said "substantially" AI-generated music isn't covered yet, and that it will extend the tag as its detection improves.

The others have stopped short. Deezer keeps detected fully AI tracks out of recommendations and editorial playlists, and calls that "a first step" to stop them diluting the royalty pool. In its 13 October announcement, it said removing or demonetising content would "need to be based on careful consideration". Spotify's 2025 policy describes AI use as "a spectrum, not a binary" and says all music is treated equally regardless of the tools used to make it. Its main defence of the pool is aimed at spam behaviour, not at AI as such.

Who decides what "purely AI" means?

This is the weak point. The phrase sounds precise, but every service draws the line in a different place, as we found when we compared AI labels across platforms. Even Tidal's tag covers "wholly" AI music today and only later will extend to "substantially" AI-generated tracks. Mehra gave no definition and no detection method on the call.

For an independent Indian artist, that matters more than it does for Saregama. A label with a large human catalogue and direct relationships with the platforms can dispute a flag. A singer-songwriter who used an AI tool to clean up a vocal, or generated a backing texture under a melody she wrote, depends on a detector she can't see and on a distributor that may not tell her anything. India's law doesn't help much yet. RMX Music, an Indian distributor, notes that the government's second working paper on AI outputs and ownership has still not been published, and advises artists to keep project files, drafts and dated voice notes as proof of their own work.

There is a real difference between a producer who sketches chords with an AI assistant in a browser tool such as Veena and then writes, plays and mixes the rest, and a script that uploads thousands of generated tracks. A fair rule has to be able to tell the two apart.

What we think

The principle is right. Streams that come from bulk-generated tracks, many of them fraudulent by Deezer's count, shouldn't take money from people who write and perform music. Deezer says up to 85% of streams on fully AI tracks in 2025 were fraudulent. Excluding that output from the pool is a reasonable defence.

But follow the money. In a shared pool, a track that earns nothing doesn't make the pot smaller. Spotify says of its own 1,000-stream threshold that excluded streams are redirected to other artists and don't increase its profit. Under pro-rata payouts, the redirected money is likely to go mostly to whoever already has the largest share of streams, and in India that means the big catalogue owners. That's our inference, not something Mehra said, but it explains why a label that reports no leakage today is still pushing hard on this.

The second-order risk is the licensed case. Mehra says Saregama will license its catalogue to AI firms. If AI output from a licensed model is later allowed back into the pool and unlicensed output isn't, "slop" stops being a quality judgement and becomes a licensing one. Tidal has already acknowledged that debate, Music Ally reported: whether music from properly licensed models should earn royalties is open, so its rule could change. Artists should push for three things before any of this becomes standard: a published definition, notice when a track is excluded, and a way to appeal that doesn't go through the label that benefits.