Appeals court in Thomson Reuters v. ROSS sees a potential AI training data market
The Third Circuit said the lack of existing licenses does not disprove the market, per a Baker Botts summary.
On Sept. 29, 2026, the Third Circuit issued what appears to be the first federal appellate decision on training AI on copyrighted material, in Thomson Reuters v. ROSS, according to a Baker Botts summary.
“the evidence shows that the market for licensing headnotes as text to train AI is rapidly developing”
Third Circuit, as quoted by Baker Botts
Under the fourth fair-use factor, the court found harm to the market for the headnotes and recognized a potential market for licensing them as training data, per Baker Botts. It held that the absence of third-party licenses does not disprove such a market. The court distinguished AI that cannot generate original expression from generative models.
Sources
Spotted an error? See our corrections policy.
Related
Australian music industry rejects Labor AI music copyright carve-out
ARIA's CEO called a reported music exemption from a planned AI training-permit regime unnecessary and unworkable. Writers' groups and Senator David Pocock also objected.
Australia weighs paid permits for AI training on local content, music likely exempt
The Australian Financial Review reports that labs such as Anthropic and OpenAI would need to buy a permit to train on Australian content under a regime the government is developing.
Australian music, screen and news groups launch 'Not AI's to Take' copyright campaign
The campaign asks the Albanese government to keep copyright law as is, so AI companies need permission and agreed payment to train on Australian work.
Character.AI gets a privacy ombudsman as states object to Disney's $20 million deal
A court-ordered privacy reviewer will weigh any sale of Character.AI user accounts and chat data. On the objection deadline, 39 states and DC asked the judge to reject Disney's stalking-horse license bid.