23 September 2026
One of the most important copyright cases involving generative AI is moving closer to a decision.
OpenAI, Microsoft, The New York Times and a group of authors have filed competing applications for summary judgment in consolidated US proceedings.
The applications ask the court to decide a central question: can copyrighted books and journalism be used to train generative AI models without permission under US fair use law?
What is the dispute about?
The New York Times and the authors allege that copyrighted works were copied without permission to train OpenAI’s models.
They argue that generative AI products compete with the original works, divert audiences and weaken the market for professionally produced journalism and writing.
OpenAI and Microsoft argue that training is transformative. Their position is that models analyse works to identify statistical patterns in language rather than reproduce the protected expression itself.
The court is now being asked to determine whether the fair use defence can resolve the central issue without a full trial.
The arguments have been made, but the court has not yet decided them.
Why the US decision will matter
A decision could significantly affect how AI developers obtain training material and how publishers, creators and other content owners license their work.
If the court accepts the fair use argument, AI companies may have greater freedom to train models on copyrighted material in the United States.
If the content owners succeed, developers may face greater pressure to obtain licences, disclose their training sources and compensate rights holders.
Either outcome could influence AI contracts and licensing negotiations well beyond the United States.
But it will not determine Australian law.
Fair use is not the Australian test
Australia does not have the broad US doctrine of fair use.
Australian copyright law instead provides specific fair dealing exceptions for purposes including:
-
research or study;
-
criticism or review;
-
reporting news;
-
parody or satire; and
-
giving professional legal advice.
Commercial AI training does not automatically fall within those exceptions.
The Australian Government has also ruled out introducing a broad text and data mining exception that would allow developers to use creators’ works for AI training without permission.
Australian businesses should therefore be cautious about treating a favourable US decision as permission to copy material for AI training here.
The relevant questions include where the copying occurred, which law applies, how the material was obtained and whether the developer had permission to use it.
What should Australian businesses do now?
AI developers should maintain clear records of their training data and the basis on which it was acquired. This includes licences, website terms, permissions and any restrictions attached to third party datasets.
Content owners should consider how their material is being accessed, whether licensing opportunities exist and what evidence would be available if unauthorised copying were suspected.
Businesses purchasing or deploying AI systems should also look beyond functionality and price. Procurement checks should address:
-
the source of training material;
-
warranties about lawful use;
-
responsibility for infringement claims;
-
the scope and exclusions of any indemnity;
-
whether customer inputs are used for further training; and
-
where relevant copying and processing take place.
A supplier saying its practices are lawful under US fair use may not be enough for an Australian customer.
The broader lesson
The US proceedings may become a major precedent for generative AI. They may also change the commercial balance between technology companies and copyright owners.
However, copyright remains territorial.
Australian businesses need an Australian legal analysis, not an assumption that a US fair use argument applies everywhere.
The safest approach is to understand what material an AI system uses, how it was obtained and what rights support that use before the system becomes commercially important.
If your business is developing, purchasing or using AI systems, IP Solved can help you assess copyright ownership, training data, licensing and contractual risk before problems arise.
