Thomson Reuters president and CEO Steve Hasker has set out the three key ingredients the news agency and business information giant has looked for when it comes to agreeing licensing deals with AI companies.

Hasker told the Truth Tellers Summit in London last week that 175-year-old Reuters has focused its deals on its text archive, while setting them at the “highest price” possible and keeping them brief in length so they will need to be renegotiated.

He said generative and agentic AI feels like it will be “more transformational and more disruptive” than the arrival of the internet, Google and social media.

“And so I think that the opportunities and the risks could be, and likely will be, even bigger as newsgatherers license or don’t license their content,” he said.

Hasker said Reuters has “taken a cautious approach with that in mind”.

“We have done a select number of deals. I can’t look anyone in the eye and tell you we’ve nailed it. But what we’ve done is we’ve basically said we’re not taking the live news file, so we’re not taking the news feed. It’s the archive, it’s not the images, it’s not the video, it’s not the audio, it really is not ‘just’, but it’s the text.

“We’ve tried to get the highest price because we want to fund the next generation of journalists and the next wave of investigations and stories and breaking news. We want to be in 200-plus countries across the world. And we’ve tried to keep the deals to be short term. And so that’s where we sit.

“I know different media outlets, different news outlets, have done different things, but we’ve tried to avoid litigation as much as we can and work sensibly and selectively with the frontier model,” he said, meaning the most advanced general purpose AI models.

By March 2024 Reuters had signed “a number” of undisclosed deals with AI companies, meaning it is unclear whether it has any agreements with the likes of OpenAI and Perplexity. But Hasker spoke about “not getting a penny” for Reuters news content appearing in major chatbots.

There appears to be a distinction between selling access to the Reuters text archive and allowing Reuters news to be surfaced and signposted to in AI news searches. Press Gazette has asked Reuters for clarification on this point.

A spokesperson said: “We do not comment on commercial agreements.”

Reuters became a publisher partner for Microsoft’s Copilot in October 2024, meaning the product can pull from its content (alongside other opted-in publishers) to give a daily AI-voiced news summary.

In the same month Reuters became the first news publisher to sign a multi-year AI deal with Meta for use of its content to answer questions relating to news and current events from chatbot users. Reuters itself reported that Meta did not disclose whether it would be able to train its large language model on the news publisher’s content.

[Read more: News agency bosses at Reuters and PA explain differing ‘red line’ on use of AI]

Hasker recalled a conversation he had with an unnamed AI boss (he denied the person in question was OpenAI CEO Sam Altman) around the unauthorised use of breaking news updates from Reuters in AI answers.

He said there is a philosophy in Silicon Valley that “if I can scrape it, it’s fair use… if I can scrape it and get hold of it, then all bets are off.

“I had a conversation many months ago now with the top of the house of one of the frontier models, and essentially said ‘look, it’s very clear to us that you’ve scraped the Reuters news file and it’s playing a significant role in fuelling the output of your chatbot’. And he said ‘no, it’s not. We haven’t done that. We’re very respectful.’”

According to Hasker, the pair were speaking the day after a “particularly vicious volley of missiles from Russia to Ukraine” so he searched for information about this on the chatbot in question and was shown what was effectively “the Reuters article in its entirety”.

He showed this to the AI leader who claimed this was a “one-off” so Hasker searched again, this time for information about the conflict between India and Pakistan. “Exactly the same thing happened,” he said.

Hasker continued: “There are still instances today where you’ll enter a query into a chatbot and it will source Reuters and we’re not getting a penny for that. And most importantly, that is not going to fund great journalism. And I think that’s a travesty.”

Thomson Reuters won a “fair use” judgment last year as a US judge ruled against a competitor using content held by its legal information provider Westlaw to train an AI tool.

Hasker noted that the “news industry doesn’t have a lot of control today”, adding: “I think what makes me nervous is okay, so a frontier model has ingested all this information. They control the user experience, therefore they control the aggregation of that audience and what’s put in front of them, and then when that particular deal comes up, what’s next?

“Now, we have faith in the value of constantly replenished and refreshed news and fact-based storytelling. It remains to be seen whether the tech industry has that same faith and whether they will continue to remunerate journalists for that work.”

Email pged@pressgazette.co.uk to point out mistakes, provide story tips or send in a letter for publication on our “Letters Page” blog