Stockholm University has referred a high-profile AI study of Chinese schoolchildren for suspected falsification. That means its striking claim that generative AI reduced classroom test performance must now pass a more basic test: whether the data supporting it can be trusted.
DN reported that the university made the referral last week to Sweden’s research-misconduct review body. Released in June, the study drew on nearly 27,000 Chinese schoolchildren. It reported that generative AI cut homework time by about 30% and lifted homework scores by 18%, while classroom test results fell by about 20% within six months. The paper has not been published in a scientific journal, yet its conclusions received wide international attention.
According to DN, an anonymous whistleblower identified patterns said to be difficult to reconcile with genuine pupil and exam data. The allegations include pupils using AI before the named tools were available, unusually weak links between results in different subjects, and no missing exam results in parts of a large dataset. Those are allegations about the evidence base, not findings of the misconduct process.
One author, Stockholm University professor David Strömberg, told DN that he welcomed the referral and that the team was trying to determine what happened. He said the researchers received the data from a Chinese authority and hold documents certifying its authenticity, but he declined to discuss the collection process further.
The central editorial point is not that the opposite conclusion about AI and learning has now been established. It is that a study used to make a forceful public claim about educational harm is being scrutinised at the level where any such claim begins: the integrity of its underlying material. The referral does not establish fabrication. But while the review is under way, it is irresponsible to present this unpublished study as decisive evidence that generative AI harms learning.
Sources
- Dagens Nyheter · Source date: 2026-11-11
Independent editorial content. WUA is not affiliated with any political party. 2026-11-11