Who owns an AI-assisted song in India? Copyright, personality rights and the one court order so far

If you write lyrics, generate the music on an AI tool and release the result, who owns it? In India the honest answer today is: you clearly own your lyrics, you may or may not own the music, and nobody but the singer owns the voice. One court has started to look at the question, and it has not answered it yet.

This is a practical guide, not legal advice. Everything below is as we read it on 24 Oct 2026. For the filing steps, see our guide to registering a song's copyright in India; for the US, PRO and distributor rules, see how to register and release an AI-assisted song.

A song is several things, and each has its own owner

Indian law doesn't protect "a song". It protects works. Under the Copyright Act, 1957, a musical work is "a work consisting of music" and excludes the words sung with it. The author of a musical work is the composer, the author of a sound recording is its producer, and lyrics are a literary work whose author is the writer. Section 17 makes the author the first owner: "the author of a work shall be the first owner of the copyright therein", subject to the Act's exceptions.

So an AI-assisted track can contain:

Keep these apart. The first Indian court to look at an AI song criticised the plaintiff for not doing so.

The clause that seems to answer it: Section 2(d)(vi)

India, unlike the US, has a statutory rule for machine-made work. Section 2(d)(vi) says that for any literary, dramatic, musical or artistic work "which is computer-generated", the author is "the person who causes the work to be created".

Read on its own, that sounds like good news for prompt writers: you caused the music to be created, so you are its author. That was exactly the argument made in Delhi this year. The trouble is that Section 13 protects only original musical works, and the Act does not say whether output from a generative model is original, or whether typing a prompt is enough to "cause" it in the Act's sense. The clause names an author; it doesn't say a protectable work exists.

What the Delhi High Court actually said

The case is Tarun Chaudhary & Anr. v. Kuldeep Meena & Ors., CS(COMM) 297/2026. We read the order of 25 Mar 2026 on the court's own server. The facts, as the order records them:

• The plaintiff bought the song "Teri Yaadon Ki Chadar Odhe" on 16 Jan 2026 from its original creator under a purchase agreement.
• The plaint admitted the song was generated through Suno "except for the lyrics". The agreement says the music was composed using Suno on 11 Nov 2025, and that the seller held a commercial licence at the time.
• He sued the defendants for infringement, sought ₹2.5 crore in damages and asked Google (YouTube) to hold back the song's revenue.

Defendant no. 1 pointed to Suno's terms, which say Suno gives no warranty that copyright exists in its output, and argued that if the seller had no copyright, the buyer could not have any either. Google cited the US appeals court ruling in Thaler v. Perlmutter, which held that human authorship is a necessary condition under US law, while accepting that case was about a work with no human input at all.

The judge accepted that Section 2(d)(vi) "would be applicable", but framed the real question as whether a plaintiff who owns the lyrics can assert rights over the "song" per se "when the entire song was admittedly musically composed by using AI tool, except for the lyrics". He noted that the plaintiff had sued over the whole song and had not claimed any violation of the lyrics on their own. Counsel also handed up a Copyright Office registration certificate for the song.

The court then:

• said it was "not clear as to whether the plaintiff can assert copyright on the song admittedly created by the AI tool";
• impleaded the Registrar of Copyrights as defendant no. 8;
• refused an ex parte ad-interim injunction "without first being, prima facie, convinced of the plaintiff's copyrights";
• listed the matter before the court on 16 Apr 2026. We could not find a later order on the case that we were able to open, so treat this as the latest position we can confirm.

LiveLaw's report of the hearing quotes the judge from the bench: "There is no judgment on AI." It also reports him asking what the copyright was actually on.

Three things to take from it. It is an interim order, not a ruling that AI music is unprotected. A registration certificate did not settle the question; the court wanted the Registrar in the room. And the lyrics were never in doubt.

What your AI tool's terms do and don't give you

Terms of service are a contract with the tool, not a grant of copyright. Suno's terms, effective 3 Sep 2026, show the gap clearly:

• On the free or Basic tier, Suno does not assign output to you, and you may use it only for "lawful, personal and non-commercial purposes".
• On Pro or Premier, Suno "hereby assigns to you all of its right, title and interest in and to any Output owned by Suno", subject to the terms' commercial-use limits.
• On every tier, "Suno makes no representation or warranty to you that any copyright will vest in any Output."

So a paid plan gets you whatever Suno has. It doesn't promise that there is anything to have, which is the exact point defendant no. 1 made in Delhi. Read your own tool's terms on the day you generate, and save a copy.

The voice is a separate question

If the track uses a cloned voice of a real singer, copyright in the composition is not the main risk. Indian courts have treated a singer's voice as part of their personality. In July 2024 the Bombay High Court barred defendants from using Arijit Singh's name, voice and likeness, and the order expressly covered AI voice models and voice conversion tools, LiveLaw reported. In September 2025 the same court protected Asha Bhosle against an AI platform offering her voice, taking the prima facie view that converting any voice into a celebrity's without permission violates her personality rights (LiveLaw).

Both are interim orders and both rest on celebrity status. We cover the gap that leaves for working singers in our report on India's AI voice-cloning cases, and the separate performer's right to object to distortion under Section 38B in our moral rights explainer. The practical point here: owning the copyright in a track gives you no right to the voice on it.

Will the government settle it?

Not soon. DPIIT's committee published Part 1 of a working paper on generative AI and copyright in December 2025, on training. The second part is meant to cover outputs, including whether they can be copyrighted. On 4 Oct 2026 Hindustan Times reported that the committee had not met since June and that the second paper had not been prepared. Until then, the Delhi case is the closest thing to an answer.

Before you release an AI-assisted song in India

• Write down, per part, who made what: lyrics, melody, arrangement, recording, vocal. Keep drafts, voice memos and project files with dates.
• Put your own human contribution into the music, not just the prompt: a sung melody, re-played parts, real edits and arrangement. The Delhi court's doubt was about a composition that was AI-made except for the lyrics.
• If you register, register each work separately and describe it truthfully. Don't claim the composition as yours if a model wrote it.
• Generate on a plan whose terms allow commercial use, and save the terms as they read that day.
• If you buy or license a song, get the assignment in writing (Section 19(1) says no assignment is valid unless it is), and ask the seller which parts were AI-made. You can't buy rights the seller never had.
• Never use a real singer's voice without written consent.
• If you sue, say exactly which right you are enforcing: lyrics, composition or recording.

What we think

The Delhi order is more useful to artists than it first looks. It reads Section 2(d)(vi) as a rule about who the author is, not a promise that every output is a protected work. That is the right instinct. If typing a prompt made you the composer of whatever a model returns, the biggest owners of Indian music would soon be whoever generates the most tracks, and the songwriters whose work trained those models would get nothing.

But the uncertainty is costing real people now. A buyer paid for a song in good faith, the Copyright Office issued a certificate, and the court still couldn't say what he owned. The Registrar is now a party; we think the Copyright Office should use that to publish what it needs from applicants who used AI, as the US Office did in its 2023 guidance, rather than wait for the stalled DPIIT paper.

For now, the safe rule is simple: you own what you made. Make the parts that matter, and keep the proof.