AI inom polisens sekretesshandläggning : En jämförande studie om kvalitet och effektivitet mellan AI och mänskliga sekretessbedömningar.

Sammanfattning

This study examines how artificial intelligence (AI) can be used in secrecy examinationwithin police work. Secrecy examination is a confidentiality assessment of the right totake part in public documents according to the Publicity and Secrecy Act (2009:400),which often refer to preliminary investigation reports. The focus is on how AI affects quality, efficiency, and processing time. The study addresses a research gap, since there arefew empirical comparisons between human and AI-based decisions in this type of work.An experimental design was used to compare three different methods: fully automatedAI redaction, AI-assisted manual redaction, and fully manual redaction. The evaluationwas based on a constructed preliminary investigation, which was assessed against anexpert-based reference. The results were measured using precision, recall, and F1-score,as well as the time required to complete the task.The results show clear differences between the methods. Fully automated AI is veryfast, but it often fails to identify confidential information. This creates a risk that sensitivedata may be disclosed. Manual redaction shows high precision, but it also misses someconfidential information. The AI-assisted method performs best overall, as it combines high precision and recall without taking more time than manual work.The study shows that AI should not be used on its own in confidentiality assessments, given the current level of development, but in the future this may change. Instead, it works best as a support tool for human decision-making. In this context, efficiency is not only about speed, but about balancing quality, time, and resources. The conclusion is that AI-assisted work is the most effective approach. AI can improvethe quality of decisions, but human control is still necessary to ensure legal certainty and protect sensitive information.