Earlier this month Donald Trump’s Department Of Justice intervened in a big copyright legal battle involving OpenAI, telling a New York court that when the AI company trained its model on millions of existing works that was clearly ‘fair use’ under American copyright law, meaning it didn’t need to get permission from any creators or copyright owners to make copies of their works.
Some of those copyright owners have now responded, explaining why Trump’s DoJ is very wrong in reaching this conclusion. And also questioning whether that conclusion was in fact reached because of the “political and financial interests” of the Trump administration, given “AI companies and their executives have been among this administration's largest donors” and “the administration is reportedly considering OpenAI’s offer of a 5% stake in the company to ease Washington pressure”.
Although this is not a music case, nearly every lawsuit filed by a creator or rightsholder against an AI company in the US swings on whether or not AI training constitutes fair use under American law, including the music industry’s lawsuits against Suno, Udio and Anthropic.
And while the specifics of each individual legal battle could impact on whether or not an AI company can rely on the fair use defence, it would be helpful for the music industry - and all the copyright industries - if there was a general consensus that AI training is not fair use except in very specific circumstances.
Trump officials previously said it was for the courts to decide whether or not AI training was fair use. But the DoJ’s intervention in the OpenAI case made it clear what decision the Trump administration thinks the courts should reach. Any conclusion that OpenAI’s training was not fair use, it said, would be “inconsistent with basic copyright law principles” and “severely hamper” the “progress of science and arts”.
Various news organisations involved in the OpenAI case, including the New York Times, have now hit back at the DoJ’s intervention in a new court filing, setting out four arguments for why the Trump administration is wrong to claim that AI training is fair use. And those four arguments are likely relevant to every ongoing US legal battle involving a copyright owner and an AI company.
DoJ ignored facts and experts
First, the DoJ’s statement to the court is not “an informed assessment of the facts in this case”, despite the government department conceding that any analysis of the fair use defence “hinges on the specific facts and uses at issue in each case”. This was in part because the DoJ submitted its statement to the court before lots of the paperwork relating to the dispute had been submitted and made public.
That includes the paperwork which quoted an executive from OpenAI shareholder Microsoft admitting that the AI industry’s use of existing content to train its models was one of the “greatest property thefts in human history” and that any fair use defence would “make a complete mockery of the idea of fair use”.
While those comments are definitely embarrassing for OpenAI, they aren’t necessarily relevant when assessing whether AI training is or is not fair use. However, the news organisations add that the paperwork ignored by the DoJ included hundreds of documents, plus deposition transcriptions from more than 70 witnesses and dozens of expert reports.
But the DoJ wasn’t interested in any of those legal arguments or expert insights, instead arguing in favour of OpenAI’s fair use position for political reasons.
DoJ’s policy pronouncements don’t impact on fair use analysis
Which brings us to the second argument. The news organisations say that the DoJ’s fair use position mainly echoes the Trump administration’s political position that “the unfettered development of AI is of utmost importance to national security”.
The enforcement of copyright isn’t actually stopping the development of AI - the news organisations argue - but even if it was, they say, the US Supreme Court has “long recognised that courts do not abdicate their judicial function merely because the executive invokes national security”.
Copyright isn’t an hindrance to AI development
The third argument delves deeper into the DoJ’s claim that forcing AI companies to secure licences from copyright owners would create a major barrier that would significantly hinder AI development.
This isn’t true, the news organisations insist, because licensing deals are possible - and are being negotiated as we speak - it’s just those deals will cost the AI companies some money. But “in any industry there is payment for raw materials” and, when it comes to AI training, that means paying for licences to access content for training.
And the companies developing AI models can afford to buy licences. Many are tech giants which are “some of the largest companies in the world”, while AI businesses like OpenAI and Anthropic are “expected to go public with stock offerings that would value each company at about $1 trillion”.
Expanding on this theme, the news organisations note that “OpenAI entered into a partnership with Microsoft in which Microsoft provided OpenAI with access to the computing power and other resources OpenAI needed”, and as a result “Microsoft has spent more than $100 billion on OpenAI’s development”.
“The need to access resources valued at $100 billion was certainly a tremendous barrier to OpenAI’s entry into the AI industry”, they go on, and if we apply the DoJ’s logic to this, “Microsoft should have been required to provide its resources to OpenAI for free”.
Plus “the fact that OpenAI needed $100 billion from one source merely to begin development of its AI models shows that, on the list of barriers to entry into the AI world, licensing fees for copyrighted content are barely a footnote”.
The DoJ ignored the Copyright Office
Finally, when a US government department intervenes in a court case, it would usually first consult any relevant government agencies. Which in this case would include the US Copyright Office.
But the DoJ didn’t do that, which is unsurprising, given the Copyright Office is still being led by Shira Perlmutter, who Trump tried to fire the day after the publication of a Copyright Office report on the big debate around AI training and fair use in May last year.
Although that report said AI training could be fair use in some circumstances, it was generally seen as favouring copyright owners more than tech companies, which means the DoJ is no fan of the Copyright Office’s conclusions.
The news organisations continue, “that the DoJ fails to acknowledge any possible contrary views, and dismisses the Copyright Office’s comprehensive study in a footnote, suggests that the DoJ may have been influenced by extraneous considerations - up to and including the administration’s political and financial interests”. Including the donations to Trump and that talk of the US government getting 5% of OpenAI.
The news organisations do concede that “political factors will always influence government litigation positions to some extent", but then insist that “the nature, extent and transparency of the AI industry’s efforts to influence this administration should not be lightly dismissed”.