Vox Media president Pam Wasserstein sent a Slack message and email to staff on May 29 detailing news that the company's journalists said was shocking: Vox had signed a content licensing deal with OpenAI.
The deal gives the AI company access to Vox's full archive of current content and journalism work to train ChatGPT and other models. Wasserstein's alert came just before Axios exclusively revealed details of the licensing and product agreement, much to the surprise of his reporters.
A reporter at The Atlantic, which has a similar deal with the Microsoft-backed AI giant, was also sent the email shortly before the Axios story was published.
“Atlantic staff learned about the contract primarily from outside sources, and the company and OpenAI have declined to answer questions about the terms of the contract,” Atlantic Union said in a May 30 statement.
Current and former journalists from both companies interviewed by TechCrunch never expected their jobs to be handed over to OpenAI, and they all worry their employers are making a shortsighted deal that will ultimately hurt writers and journalism as a whole.
Vox Media (which includes The Verge, New York, Eater, The Cut, and others) and The Atlantic have published articles critical of OpenAI and generative AI. Amy McCarthy, a reporter for Eater and communications chair for Vox's union, said the articles raise concerns about the environmental impact of the electricity needed to run large language models, turmoil on OpenAI's board, and the company's “general lack of credibility.”
Vox did not respond to a request for comment.
Since the deal was announced, journalists at each publication have been meeting with business executives to learn more about the deal, and they've been asking for one key piece of information: What's in it for journalists?
Sense of urgency
Faced with a rise in AI media deals, the news industry is accelerating negotiations to put in place AI protections similar to those fought for by Hollywood writing teams.
“The Writers Guild and the Vox Media Union firmly believe that even if there's no AI clause in the contract, the introduction of AI is a must-have for negotiation,” McCarthy told TechCrunch. “The contract basically contains a clause that says the company has to negotiate with us about fundamental changes to working conditions. We strongly believe that this is a workplace issue, this is a working conditions issue, and that the company has an obligation to negotiate with us about how this works.”
This means that publishers who enter into agreements with AI providers may be contractually obliged to consult and negotiate these changes with trade unions.
The Atlantic Media Union also intended to bring the issue to the negotiating table, but the OpenAI deal added urgency to the effort, said a current employee who spoke to TechCrunch on the condition of anonymity.
In negotiations this month, The Atlantic's union put forward a proposal not to use AI as a substitute for writing, fact-checking, copyediting or illustrations, and that reporters could use AI at their own discretion and in accordance with journalistic principles and ethics, but could not be forced to do so. The proposal has yet to be accepted.
Other unions are working to secure similar protections: Nebraska journalists at the Omaha World-Herald Guild secured protections from AI earlier this year. After CNET published a series of AI-generated stories in 2023, journalists at the paper went public with their union drive, demanding AI protections and a say in how AI is implemented in employee workflows.
Getting companies to include such safeguards in their contracts with journalists is crucial, because the law doesn't guarantee protection. Companies like OpenAI argue that they don't violate copyright law when they scrape what they say is public content, and that their chatbots don't copy material in its entirety.
However, publications including The New York Times, Raw Story, Alternet and The Intercept have sued OpenAI for using journalists' work to train ChatGPT without properly crediting or citing it. Novelists, computer programmers and other groups have also filed copyright lawsuits against OpenAI and other companies building generative AI.
Richard Tofel, former president of the nonprofit news outlet ProPublica and a news organization consultant, believes these cases will ultimately end up in the Supreme Court: If the court finds OpenAI or others guilty of copyright infringement, “they'll have to settle with everybody.”
Tofel believes most publishers will eventually strike deals with AI companies, noting that Google faced a similar copyright lawsuit when its search product was first launched, but by the time it was resolved, users had become so reliant on search that no publisher wanted to exclude their content.
McCarthy said authors cannot rely solely on the courts: “We need to explore all options for fighting the introduction of AI.”
Another concern for journalists is publishers' adoption of AI in content creation, with some media outlets already starting to experiment.
CNET and Gannett have published AI-generated articles and art, and in the case of Sports Illustrated, fabricated bylines. These articles have been accused of being AI-generated primarily because they were riddled with factual errors, but these obvious errors may decrease over time if the AI is allowed to freely learn good journalism.
If journalists don’t question this, who will?
Journalists understand the basic structure of the deal, but still have questions.
Anna Bloss, vice president of communications for The Atlantic, said the partnership, like other publisher deals, positions the company as a premium news source within OpenAI.
“Articles from The Atlantic will be discoverable within OpenAI's products, including ChatGPT. As a partner, The Atlantic will help shape how news is displayed and presented in future real-time discovery products,” Bross told TechCrunch. “This agreement ensures guardrails and protections around how our content appears within OpenAI's products. … If an Atlantic article appears in response to a query, it will have Atlantic branding and a link back to the article on our site.”
Bross pointed out that this is not a syndication license, which means that OpenAI does not have permission to reproduce The Atlantic articles or to make similar copies of the entire articles or longer excerpts.
But Atlantic journalists are still waiting for an explanation from management as to why such content doesn't qualify as derivative work, which would give them a chance to be paid directly: The paper recently launched a new series of paperbacks featuring work by its writers and has paid them for the work, multiple sources told TechCrunch.
Bross noted that The Atlantic's contract with OpenAI prohibits the creation of derivative works.
The Atlantic's editorial staff raised the topic at an all-hands meeting led by the magazine's CEO, Nick Thompson, in mid-June and learned that while ChatGPT has access to their articles, the editorial team is otherwise “fairly insulated.”
In other words, there is no imminent threat of ChatGPT being used to write articles.
Financial terms of the Atlantic-Vox deal have not yet been disclosed to journalists inside or outside of either publication, but the Atlantic's deal is for two years, and both publications are known to use OpenAI's technology to build products and features. OpenAI said its technology will not be used to mimic the voices of its writers themselves.
News Corp, the parent company of The Wall Street Journal, also signed a five-year deal with Open AI reportedly worth more than $250 million, and Axel Springer, which runs Politico and Business Insider, has also teamed up with Open AI in a deal reportedly worth tens of millions of euros.
Other media outlets that already have similar partnerships with OpenAI include Dotdash Meredith (publisher of People, Better Homes & Gardens, Allrecipes, Investopedia, etc.), The Associated Press, the Financial Times, France's Le Monde, and Spain's Prisa Media.
(It's also worth noting that TechCrunch's parent company, Yahoo, is also working on AI through its Yahoo News app, which uses the underlying code from Artifact, an app Yahoo acquired in April.)
OpenAI claims the deal will drive traffic back to articles and help journalists, but implementation has yet to begin, so it remains to be seen how well it will perform.
Tofel said that if users were to be able to ask an AI chatbot for the latest updates on the war between Israel and Hamas, for example, it would be “a news organization's ultimate nightmare.”
“AI news products have the potential to remove the middleman to a large extent,” he said.
OpenAI was unable to confirm any user experience design details that might determine the likelihood of a reader clicking on an external link to an article.
Also, advertising revenues would fall if readers no longer had to go to publishers' websites to read articles — a problem the news industry already faces after Google and Meta deprioritized news in their algorithms — and journalists and writers would find their work less readable.
With journalism suffering from a lack of funding, mainly because big tech companies such as Meta and Google now capture the vast majority of digital advertising revenues, publishers would surely welcome any new revenue stream to shore up their balance sheets.
But journalists are questioning whether this is the best way forward.
“It's like a protection racket,” McCarthy said. “It's like making a deal with a guy who robbed our house and pinky promising him he's not going to rob our house again.”
Some AI startups are already stealing content without any contracts in place. For example, ChatGPT rival Perplexity has been accused of plagiarism by Forbes, and Wired recently found that the AI company had been secretly scraping the publisher's website. Despite these allegations, Perplexity is preparing to announce ad revenue-sharing deals with publishers next week, the startup told TechCrunch.
Still, we can expect to see more of these deals in the future, as publishers all seem to come to the same conclusion: AI is going to steal our work anyway, so we might as well get paid.
Fix: This story originally misstated how The Atlantic told staff about the deal — the email to staff was sent just before publication — and has been updated to clarify that Atlantic writers are protected from derivative works by OpenAI.
