GenAI Catch-22: Norwegian newsrooms' AI dependence erodes human oversight

A new arXiv study from researchers Mari Reisjå and Anders Sundnes Løvlie examines how Norwegian newsrooms adopted generative AI during the 2025 parliamentary election campaign. Based on interviews with managers and journalists over a ten-month period, the paper reveals a gap between ambitious AI visions and practical reality. Newsroom managers shared what the authors call “sociotechnical imaginaries”—unrealistically optimistic beliefs about AI capabilities and development speed. These beliefs fueled plans for audience-facing GenAI services, but when technological and practical challenges hit, those plans collapsed and gave way to far more mundane internal uses, like drafting memos or assisting with routine reporting tasks.

The study also reframes the dominant threat narrative. While much attention focuses on external disinformation, the researchers highlight an internal threat: journalists' own use of GenAI. This creates what they term the “GenAI Catch-22.” To monitor AI outputs and prevent errors or undue influence, newsrooms rely on human expertise. But extensive GenAI use risks deteriorating that expertise—journalists lose the sharp critical skills needed to catch subtle AI failures. The more newsrooms lean on AI, the less capable they become at overseeing it. This vulnerability is especially dangerous during high-stakes coverage like elections, where accuracy and editorial judgment are paramount. The paper, published on arXiv (2608.10773), offers a cautionary tale for any organization deploying GenAI in knowledge-critical roles.