Facts
The Municipality of Trento, with the Fondazione Bruno Kessler acting as its processor, took part in two EU-funded research projects, MARVEL and PROTECTOR, which used AI to detect public safety events such as gatherings, assaults and bag snatchings. For MARVEL it took 309 hours of footage from 14 urban security cameras and 85 hours of audio from 6 microphones placed on public streets. PROTECTOR used footage from 4 cameras and collected religious hate messages from Twitter and YouTube; the Garante found that usernames in one Twitter network were shared in clear with the Antwerp police and Bulgaria's interior ministry. The municipality said the data were anonymised immediately after collection. The Garante opened its inquiry after press reports.
What was the legal question in Comune di Trento (MARVEL and PROTECTOR projects)?
The Garante first had to decide whether blurring faces and number plates, and altering speakers' voices, actually anonymised the data. It then asked whether the municipality had a sufficient legal basis under Articles 6, 9 and 10 GDPR and Articles 2-ter, 2-sexies and 2-octies of the Italian Data Protection Code for processing that also covered criminal-offence data and religious beliefs, and whether urban security legislation, the municipality's general statutory functions or the rules on scientific research could supply one. It also examined the privacy notices, the sharing of data with project partners and the data protection impact assessment.
What did the Italian Data Protection Authority (Garante per la protezione dei dati personali) decide?
The Garante held that the techniques did not anonymise the data: voice conversion kept the content of conversations, and blurred people could still be singled out by clothing, build, location or their movement between cameras. The general provisions relied on were not clear and foreseeable enough, urban security cameras could not be repurposed, and research was not among the municipality's functions. The notices were incomplete and social media users were told nothing. The impact assessment ignored necessity and proportionality, such as why a simulated urban setting would not do. Finding breaches of Articles 5(1)(a), 6, 9, 10, 13, 14 and 35 GDPR, it imposed a €50,000 fine, banned further processing and ordered deletion.
Why does Comune di Trento (MARVEL and PROTECTOR projects) matter?
The decision shows that a public body training AI models cannot rely on a lighter legal basis merely because the processing serves research or technology development. Drawing on Glukhin v. Russia and CJEU case law, the Garante stressed that conduct in public places can still fall within private life. It also shows that common techniques such as blurring and voice conversion do not by themselves amount to anonymisation, and that an impact assessment cannot be confined to information security.
Related stages
The Garante's press release followed on 25 January 2024. A charge under Article 28 GDPR concerning the processor's appointment was dropped after the municipality's submissions. According to L'Adige (10 April 2024), the municipality paid the reduced amount of €25,000 under Article 166(8) of the Code. The decision records that no data had yet been processed in a third project, PRECRISIS. On surveillance in public spaces, see glukhin-v-russia and r-thompson-carlo-v-met-police-live-facial-recognition.