LIETUVIŲ · TRANSLATION NOT PUBLISHED YET
This page has no approved Lithuanian translation, so you are reading the English original. It is not listed in search results in this language. OPEN THE ENGLISH ORIGINAL ↗
AI
Can AI train on my music?
In the EU, AI companies can train on your music unless you have opted out using a machine-readable reservation, under the text and data mining exception in the Copyright in the Digital Single Market Directive. In the US, there is no equivalent opt-out right in statute, and companies are instead relying on unsettled fair use arguments that courts are actively deciding case by case. Either way, an unauthorised use for training is not automatically legal just because a model was trained on public internet data.
The EU opt-out mechanism
The EU's text and data mining exception allows AI developers to mine copyrighted works, including music, unless the rightsholder has expressly reserved their rights, in a way readable by machines, such as through metadata or a robots.txt style declaration. This is often called a TDM opt-out. If you have not opted out anywhere your music is hosted or distributed, EU-based training may be lawful even without your direct consent.
Opting out is not automatic. You, your distributor, or your platform needs to actively implement the reservation. Many independent artists have never checked whether their distributor does this on their behalf.
The US position is unsettled, not settled
US copyright law has no dedicated TDM exception. AI companies argue that training is transformative fair use. Rightsholders argue that ingesting entire catalogues to build a commercial product is not transformative in the way the doctrine intends. Multiple lawsuits are working through US courts and the outcome will likely differ by fact pattern, so treat any blanket claim that AI training is "legal" or "illegal" in the US as premature.
What you can do regardless of the legal outcome
You do not have to wait for a court ruling to protect your position. Add explicit training restriction language to your distribution and licensing agreements, implement machine-readable opt-out signals where your platform supports it, and keep a dated, verifiable record of when you released each track and under what terms.
This record matters most if you later need to prove your work existed and was restricted before an AI company's training data was assembled.
KEY FACTS
- 01The EU TDM exception permits training unless a valid machine-readable opt-out exists.
- 02The US has no statutory TDM exception; the debate centres on fair use.
- 03Opt-outs must generally be implemented by you, your label, or your distributor, not assumed.
- 04Multiple active US lawsuits will shape how courts treat AI training on copyrighted audio.
FAQ
Does copyrighting my song stop AI training automatically?
No. Copyright registration establishes ownership, but stopping training requires a separate opt-out or contractual restriction.
Can I sue an AI company for training on my music without permission?
You can in principle, but outcomes depend heavily on jurisdiction and the specific facts, and current case law is not yet settled.
Do streaming platforms train AI on uploaded music by default?
Policies vary by platform, so check the terms of service of your specific distributor and streaming service directly.
Is sampling my song into a new AI model the same as training on it?
They raise different legal questions; training concerns data ingestion, while output sampling concerns derivative works and infringement.
LAST REVIEWED 2026-08-01 · LIETUVIŲ