(Bloomberg) -- The US Justice Department is closely monitoring deals that artificial intelligence companies like OpenAI have been reaching with content creators, said Jonathan Kanter, the agency’s top antitrust official.

“We’re watching all these deals very carefully,” Kanter said Thursday in an interview on the sidelines of a conference at Stanford University. “Just because companies enter into deals where another side is exercising monopsony power, doesn’t mean that it’s acceptable under the law.”

In remarks at the conference earlier in the day, Kanter said AI companies need to “adequately compensate creators for their works” to prevent other industries from crumbling in the same way that journalism has since the creation of the internet.

Under Kanter, antitrust enforcers have focused more on markets where there are only a few or one buyer, known as a monopsony. In its case blocking Penguin Random House from buying Simon & Schuster, the agency successfully argued that the deal would give the combined publishing house too much power over authors’ pay.

Companies including OpenAI, Meta Platforms Inc. and Alphabet Inc.’s Google have raced to gather massive troves of data pulled from articles, books and online comments to teach AI systems to understand a user’s query. That has drawn criticism from copyright holders, who argue the companies are using their work without compensation or in ways it wasn’t intended.

OpenAI came to prominence in 2022 after the release of its ChatGPT chatbot. The startup is backed by Microsoft Corp., which invested $13 billion and has infused OpenAI’s technology into its software products. 

The company has reached a number of agreements with news providers including Rupert Murdoch’s News Corp., the Financial Times and Dotdash Meredith, owner of titles such as People magazine and Investopedia. Other OpenAI deals have included Reddit Inc., the Associated Press, Axel Springer and Le Monde.

Read More: US Justice Department Steps Up Focus on Competition in AI

While more outlets are signing on to work with OpenAI as it seeks content to train its AI systems, some publishers have resisted. The New York Times sued the company, alleging use of copyrighted materials.

OpenAI said in September that creators could apply to have copyrighted works removed from the training data for its Dall-E image generator. The process was criticized for requiring artists or copyright owners to submit an application for every image and would only apply to future models.

Meta has also been using Instagram images and videos to train its model, but only allows opt-outs for citizens in the European Union.

--With assistance from Shirin Ghaffary and Rachel Metz.

©2024 Bloomberg L.P.