The US Authorities has informed a court docket that AI firms don’t break copyright regulation once they practice their fashions on written work with no license.
The Division of Justice set out that place on Tuesday (September 1), in a submitting within the copyright lawsuit introduced towards OpenAI by The New York Occasions.
It seems to be the primary time Washington has intervened in any of the copyright circumstances now stacked up towards AI firms.
These circumstances embody the lawsuits filed towards Anthropic, Suno, and Udio by the world’s largest music firms.
Each one in every of them activates “truthful use,” the exception in US copyright regulation that permits copyrighted materials to be reused with out permission.
The DOJ has now come down on the AI business’s facet of that query – at the least, that’s, with regards to copyrighted textual content.
The DOJ’s submitting – a ‘Assertion of Curiosity Of The USA’, which you’ll be able to learn right here – is recommendation somewhat than a ruling, and Choose Sidney Stein is free to disregard it within the OpenAI case.
The 20-page doc was signed by Stanley Woodward, the Affiliate Lawyer Common.
It reads: “The USA has a powerful curiosity on this Court docket rejecting any argument that coaching LLMs on copyrighted texts violates copyright regulation.”
“The United States has a powerful curiosity on this Court docket rejecting any argument that coaching LLMs on copyrighted texts violates copyright regulation.”
Assertion of Curiosity Of The USA
The submitting rests on Donald Trump‘s personal AI coverage, citing two of the President‘s govt orders, from January 2025 and June 2026.
It additionally quotes his Nationwide Coverage Framework for Synthetic Intelligence, printed in March, which states that the “coaching of AI fashions on copyrighted materials,” in and of itself, “doesn’t violate copyright legal guidelines.”
The DOJ takes on the 2 questions that resolve most truthful use rulings: (i) how far the brand new use transforms the unique, and (ii) whether or not it damages the marketplace for it.
On the primary, the temporary argues that copying textual content (just like the New York Occasions’) to coach a mannequin like ChatGPT is “a use of a unique variety or character,” and “terribly transformative.”
On the second, the Justice Division argues {that a} coaching copy doesn’t “function an alternative to the unique,” as a result of coaching “doesn’t reveal something to the general public in any respect.”
Giant AI firms paying licensing charges to publishers of titles just like the NYT “would disproportionately profit legacy media retailers as a result of sheer quantity of their written publications,” the US Authorities provides.
It isn’t within the public’s curiosity, the DOJ argues, for the most important tech firms to carry “an oligopoly on LLM coaching as a result of licensing entry boundaries that perform primarily as massive subsidies for previous mainstream media firms.”
Why this issues for music
To be very clear: the DOJ‘s submitting is about phrases, not songs.
It argues about “copyrighted texts,” “written works,” and “textual content articles.” In the meantime, a footnote limits the DOJ‘s reasoning to this case and, particularly, associated fits introduced by “ebook authors and publishers.”
Recordings and compositions go fully unmentioned throughout its 20 pages.
However truthful use is truthful use. And, clearly, a decide weighing Suno‘s protection could learn what the US Authorities now says the check means.
The DOJ splits the constructing of an AI mannequin into three phases: (i) buying the fabric, (ii) coaching the mannequin on it, and (iii) producing outputs.
“Every stage could current distinct questions of copyright regulation,” the DOJ says – and it defends solely the center one.
Of their banner circumstances towards AI firms, the majors and their publishers are attacking all three.
The first stage is how the fabric was obtained, and it’s an space the place the AI business has already misplaced floor.
Within the precedential ebook authors’ case towards Anthropic, Choose William Alsup dominated in 2025 that downloading books from pirate libraries was not truthful use, calling it “easy piracy however at large scale.”
Anthropic settled with these authors for $1.5 billion in September 2025 over the identical torrenting.
Two of the 4 counts in Sony Music Publishing and Warner Chappell Music’s new go well with towards Anthropic, the fifth music copyright case towards the Claude developer, concern torrenting.
The DOJ‘s submitting says nothing about any of that.
The second stage at query in AI circumstances is the coaching itself (i.e. fashions being fed info/content material, and studying from it).
It’s this stage the DOJ defends, and the one place it goes straight at music’s reasoning.
In 2025, ebook authors who had sued Meta over AI coaching misplaced on truthful use. However the decide who determined it, Vince Chhabria, raised a concept that might assist rightsholders in future circumstances.
Chhabria steered that AI outputs carry the “potential to flood the market with competing works” – and that builders ought to subsequently “usually must pay copyright holders for the proper to make use of their supplies”… even for coaching.
In different phrases: for Chhabria, what comes out is proof that what went in ought to have been licensed.
Legal professionals name that market dilution, and it’s the argument music has been constructing on ever since.
In a short filed on March 30, the RIAA, NMPA, A2IM, SoundExchange, and 4 different teams requested a court docket to reject Anthropic‘s truthful use protection (in a authorized battle with UMG, Harmony, and ABKCO) on comparable market hurt grounds.
Nevertheless, the DOJ now calls Chhabria‘s reasoning “deeply flawed,” and says he “improperly collapsed LLM coaching and LLM outputs right into a single steady use.”
Coaching and outputs are two separate authorized questions, the US Authorities argues, and what a mannequin produces has no bearing on whether or not coaching it was lawful.
If a court docket accepts that, music can not level at a flood of AI tracks as proof that coaching on UMG or Sony recordings was illegal.
That wall cuts each methods, which brings us to the third stage of the ‘AI wars’: what the fashions really places out.
The DOJ just isn’t defending outputs – it’s saying they should be fought over individually.
On the output stage, the US Authorities concedes, “sure makes use of will not be transformative if the LLM reconstructs and disseminates an unique copyrighted work.”
As MBW reported in July, that’s the floor UMG and Sony Music have chosen towards Suno and Udio: that AI-generated songs compete instantly with the recordings used to coach the fashions that made them.
Music publishers make the identical argument about Claude reproducing lyrics on demand. (A fourth declare sits outdoors truthful use altogether: Sony Music Publishing and Warner Chappell accuse Anthropic of stripping out copyright administration info, the possession knowledge connected to a piece.)
The New York Occasions mentioned on Wednesday (September 2) that the Trump administration “is siding with a handful of trillion-dollar AI firms on the expense of the numerous American creators whose work they stole.”
“Each AI and creators can thrive – AI firms merely must pay pretty for the content material that makes their merchandise doable, as copyright regulation requires,” mentioned Graham James, a spokesperson for the paper.
“The Administration’s proposal to let firms take that content material with out permission or compensation would undermine the sustainability of the human-created content material {that a} wholesome society will depend on, and which AI must perform.”Music Enterprise Worldwide
