Three days in Copenhagen and the lingering thought: news may not be the asset we think it is.

Shuwei Fang’s and David Caswell’s Signals at Scale brought together news, civil society, and academia to think from the ground up what an AI mediated news ecosystem can and should look like.

“Be suspicious of solutions that require the least amount of change,” must by now be a famous Shuwei quote.

After torturing our brains for a day and a half, it was time for the fourth Nordic AI in Media Summit.

It’s the sort of conference where in between sessions you can talk strategy with CEOs and moments later someone shows you live agents on their laptop. Organizers Agnes Stenbom Swedling, Kasper Lindskow, Olle Zachrison and Sara Inkeri Vardar made sure to combine big questions with real-world use cases – to the ominous sounds of Video Killed the Radio Star.

Ezra Eeman (NPO): most newsrooms stay close to the shore with AI, while journalism is moving into a new territory with the agentic web.

Nikita Roy (Newsroom Robots Labs): if you knew nothing about websites, only that people need verified information to navigate their lives, what would you build?

Florent Daudens (Mizal) urged publishers to capture AI agents, to own the demand signal, explore where the value lives, and collect payments.

Rasmus Kleis Nielsen (University of Copenhagen): content gets commoditized, connection does not. If news is not the asset the giants want, the move is not to defend the article. It is to expand into what is genuinely hard to reproduce.

Use cases

VG (Schibsted) runs its app VG X fully autonomously: editors do not approve every output, they steer by requesting changes inside the product. Human on the loop, not in it. Editor-in-chief Gard Steiro‘s slide read “100% AI NO HUMANS,” but he actually said: double down on the human touch AI cannot replace.

Also from Schibsted: Videofy, now open-sourced, turns an article into video in about 15 minutes. (Juan Carlos Lopez Calvet, Josefine Ytre-Eide Bjaarstad).

Bonnier’s editorial assistant Amelie reached 74% of journalists and 63,000 chats a month (Lina Hallmer).

Aftonbladet and Expressen built World Cup chatbots grounded in verified articles, not vibes (Angelica Öhagen, Jakob Wagner).

The New York Times aimed AI at the Epstein files: 3 million pages, 11 years of reading for one human. Reverse image search was the biggest unlock (Zach Seward, Rubina Fillion).

Olga Robinson (BBC Verify) and I dug into verification. No silver bullet, probably never one, and the deepfakes are now good enough that people do not recognize fakes of themselves.

Low-tech trick (for now): open PowerPoint, trace the perspective lines in a suspect image, and see if they converge on one vanishing point. If they do not, it is probably AI.

(also on LinkedIn)