PrevBOT — Norway's Police University College Tests Whether Grooming-Detection AI Can Lawfully Be Researched, With Mass-Surveillance Warnings

PrevBOT is a concept-stage research project at the Norwegian Police University College (PHS) to build an AI tool that scans open internet chat arenas for sexual exploitation of minors. The plan was to train it on chat logs from criminal cases where courts established that grooming took place, so police could identify arenas needing vigilance and flag conversations.
The project entered Datatilsynet's regulatory sandbox for AI; the exit report was published on 20 March 2024. It concluded that PHS has assessed and outlined how it can lawfully conduct research into such a tool, and a preliminary ethical analysis found it ethical to begin research on criminal-case data. It warned of a risk of 'the ends justifying the means' as the project advances, recommended identifying unwanted uses early, and cautioned that a green light for research does not necessarily cover lawful operational use, suggesting a further sandbox project on that question. PHS planned a Tsetlin machine chosen for explainability; the sandbox assessed this only in theory and did not test it.
An independent academic study of the sandbox (Paul and Åm, Cambridge Forum on AI: Law and Governance, December 2025) reports that the regulator was 'very vocal about the lack of a legal basis' for processing live chatroom data to train a model, and quotes an official noting the tool 'has also the potential of being a mass surveillance tool'. Earlier Norwegian-language chat research cited in the project's academic papers reached a best accuracy of 89%.
This is a cautionary and early-stage case: no operational deployment, no measured outcomes, and non-binding guidance only.

Read the full analysis: https://www.scup.com/doi/pdf/10.18261/issn.2703-7045-2021-02-01