The US Copyright Office has thoughts on how AI is trained. Big Tech may not like it.

Big Tech companies train their AI models mostly on the work of other people, like scientists, journalists, filmmakers, or artists.

Those creators have long objected to the practice. Now, the US Copyright Office appears to have joined their side.

The office released on Friday its latest in a series of reports exploring copyright laws and artificial intelligence. The report addresses whether the copyrighted content AI companies use to train their AI models qualifies under the fair use doctrine.

AI companies are probably not going to like what they read.

AI companies are desperate for data. Most of them believe that the more information a model can digest, the better it will be. But with that insatiable consumption, they risk running afoul of copyright laws.

Companies like Open AI have faced a slew of lawsuits from creators who say training AI models on their copyrighted work without permission infringes on their rights. AI execs argue they haven’t violated copyright laws because the training falls under fair use.

According to the US Copyright Office’s new report, however, it’s not that simple.

“Although it is not possible to prejudge the result in any particular case, precedent supports the following general observations,” the office said. “Various uses of copyrighted works in AI training are likely to be transformative. The extent to which they are fair, however, will depend on what works were used, from what source, for what purpose, and with what controls on the outputs — all of which can affect the market.”

The office made a distinction between AI models for research and commercial AI models.

“When a model is deployed for purposes such as analysis or research — the types of uses that are critical to international competitiveness — the outputs are unlikely to substitute for expressive works used in training,” the office said. “But making commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets, especially where this is accomplished through illegal access, goes beyond established fair use boundaries.”

In the report, the office compared artificial intelligence outputs that essentially copy its training materials to outputs with added elements and new value.

“On one end of the spectrum, training a model is most transformative when the purpose is to deploy it for research, or in a closed system that constrains it to a non-substitutive task,” the office said. “For example, training a language model on a large collection of data, including social media posts, articles, and books, for deployment in systems used for content moderation does not have the same educational purpose as those papers and books.”

Training an artificial intelligence model to create outputs “substantially similar to copyrighted works in the dataset” is less likely to be considered transformative.

“Unlike cases where copying computer programs to access their functional elements was necessary to create new, interoperable works, using images or sound recordings to train a model that generates similar expressive outputs does not merely remove a technical barrier to productive competition,” the office said. “In such cases, unless the original work itself is being targeted for comment or parody, it is hard to see the use as transformative.”

In another section, the office said it rejected two “common arguments” about the “transformative nature of AI training.”

“As noted above, some argue that the use of copyrighted works to train AI models is inherently transformative because it is not for expressive purposes. We view this argument as mistaken,” the office said.

“Nor do we agree that AI training is inherently transformative because it is like human learning,” it added.

A day after the office released the report, President Donald Trump fired its director, Shira Perlmutter, a spokesperson told Business Insider.

“On Saturday afternoon, the White House sent an email to Shira Perlmutter saying ‘your position as the Register of Copyrights and Director at the US Copyright Office is terminated effective immediately,” the spokesperson said.

While Trump, with the help of Elon Musk, who has his own AI model, Grok, has sought to reduce the federal workforce and shutter some agencies, some saw the timing of Perlmutter’s dismissal as suspect. New York Rep. Joe Morelle, a Democrat, addressed Perlmutter’s firing in an online statement.

“Donald Trump’s termination of Register of Copyrights, Shira Perlmutter, is a brazen, unprecedented power grab with no legal basis. It is surely no coincidence he acted less than a day after she refused to rubber-stamp Elon Musk’s efforts to mine troves of copyrighted works to train AI models,” the statement said.

Big Tech and AI companies have rallied around Trump since his election, led by Musk, who became the face of the White House DOGE Office and the administration’s effort to reduce federal spending. Other tech billionaires, like Meta CEO Mark Zuckerberg and OpenAI CEO Sam Altman, have also cozied up to Trump in recent months.

A representative for the White House did not respond to a request for comment from Business Insider.

The post The US Copyright Office has thoughts on how AI is trained. Big Tech may not like it. appeared first on Business Insider.