Who is winning the AI training copyright fight?
Do AI developers need permission to train on copyrighted work, and how are courts, governments and rights holders answering that in 2025-26?
Courts have split on whether models contain copies, and the money is moving through settlements and licences rather than verdicts. Governments are diverging too: the UK has shelved a training exception while the US executive branch now argues training is fair use.
- Rule · In forceEUGPAI Code of Practice
The EU's GPAI code sets the baseline expectation that model providers respect opt-outs and limit infringing outputs.
- UK High Court rejects Getty's secondary copyright claim against Stability AI
London holds model weights store no copies, narrowing claims against models trained abroad.
- Munich court rules ChatGPT's memorised song lyrics infringe copyright (GEMA v OpenAI)
Munich reaches the opposite view on memorised lyrics, treating weights as reproductions.
- Warner Music settles with Suno and agrees licensed AI music models for 2026
The music majors trade lawsuits for licensed models and output controls.
- Rule · In forceGBUK Copyright and AI report
The UK abandons its preferred opt-out training exception and waits for more evidence.
- Court gives final approval to Anthropic's $1.5bn settlement with book authors
A $1.5bn settlement prices the use of pirated books for training.
- Delhi High Court refuses ANI's bid to stop OpenAI training on its news
India's first ruling provisionally treats training on news as fair dealing.
- US Justice Department backs OpenAI's fair-use defence in New York Times case
The US government sides with OpenAI as the Times case reaches summary judgment.
- Third Circuit upholds ruling that ROSS's AI training on Westlaw was not fair use
The first appellate ruling rejects fair use where the AI product competes with the source.
Where this line could go next
Connected developments not on this line
- DevelopmentEU publishes General-Purpose AI Code of Practice ahead of AI Act model duties10 Jul 2025 · Rule change · EU
- DevelopmentOpenAI ties large reasoning-distillation campaign to people linked to Moonshot AI30 Sep 2026 · Incident · US, CN
- DevelopmentOpenAI launches GPT-6 Astra, first model it rates 'Critical' for cyber capability3 Sep 2026 · Model release · US
- DevelopmentTrump executive order targets state AI laws with litigation and funding pressure11 Dec 2025 · Policy · US
- DevelopmentEBU-BBC study finds almost half of AI assistant news answers have a significant flaw21 Oct 2025 · Research · INTL
- DevelopmentWhite House sends Congress AI framework calling for preemption of state laws20 Mar 2026 · Policy · US