Italy has reportedly given its police a legal basis for real-time facial recognition in public spaces. One caveat first: the decree's identity rests on secondary reporting. It is described as Legislative Decree No. 160 of 9 September 2026, in force from 30 September. We could not retrieve the Gazzetta Ufficiale text, so the details below come from that reporting and may differ from the final instrument.
The case for the decree
The strongest argument for it is that the alternative is worse. The EU AI Act (Regulation (EU) 2024/1689) generally bans real-time remote biometric identification in public spaces for policing. It leaves narrow exceptions, such as finding missing persons, preventing threats to life and stopping terror attacks, as EUobserver's summary of the rules puts it. The EU's Fundamental Rights Agency (FRA) found that some member states have not yet set up the legal frameworks or supervisory authorities that these tools need. Where the law is silent, police can use the technology on an ad hoc basis with no rules and no one watching. A statute with defined purposes, deletion duties and oversight is better than that.
What the decree reportedly does
According to one detailed account, the decree:
- limits real-time identification to specific, imminent threats to life or physical safety, genuine terrorist threats, and searches for missing or abducted people;
- bans building biometric archives by scraping images from the internet;
- requires comparison databases to be built fresh for each use and deleted when the authorisation expires;
- requires qualified human review before any AI output affects a person's legal position, plus tamper-proof logs and fundamental-rights impact assessments;
- puts a preliminary-investigations judge, the prosecutor and the data protection authority in an oversight role, with a 15-day, renewable activation period.
That design is what proportionate regulation looks like. The purposes are narrow, the data lifetime is short and a person has to be accountable for each use. Italian authorities have also kept AI-assisted identification to a defined set of uses and left out blanket monitoring. The same account says the decree "does not authorize indiscriminate mass surveillance."
Where the risk sits
The weak points are the emergency routes and the retention rules. The same account describes an urgent route where a prosecutor activates the system and sends the order to a judge within 24 hours. In extreme urgency, judicial police can switch it on themselves and notify the prosecutor within 12 hours. Emergency powers meant as exceptions tend to become routine. The FRA warns that measures introduced as exceptional can become standard over time, especially under internal-security policies.
The July draft also drew criticism. Eunews reported that it would allow biometric data of people passing sensitive locations such as squares, stadiums or public events to be stored for seven days so it could be compared if a crime occurred. The data protection authority opposed the broad definition of those locations. Our reporting does not show whether that provision survived into the final text. If it did, it would cut against the "event-specific, then delete" logic that the decree is credited with. The Commission's response then was general: a spokesperson said the AI Act is "a clear prohibition" on monitoring people's daily movements in accessible public spaces, and did not rule on the Italian text.
Assemblies are the test
The FRA's report, published on 30 September 2026, covers seven member states, Italy among them. Its survey found that 37% of respondents fear their data could be used for unauthorised purposes. It also found that 29% would be less willing to join peaceful protests if facial recognition were used. Those figures describe a chilling effect. That harm to free expression can arise even if every match is accurate. The FRA recommends that real-time use not target people at peaceful demonstrations.
This is where a pro-speech reading of the decree matters. A rule limiting use to terror threats and missing persons is defensible. But a squares-and-stadiums retention window, or a loosely read "specific threat" ground, would put protest crowds inside the system. For a free-expression publication, that is the line that counts.
What should happen next
- Publish the decree's impact assessments. The FRA says authorities should go beyond the AI Act and publish the results of their fundamental-rights impact assessments. Italy should do so.
- Report on emergency activations. Publish counts of urgent and extreme-urgency activations, and how often judges refused to validate them. If emergency routes dominate, the safeguard is not working.
- Confirm that retention is event-specific. If any blanket retention of passers-by's faces exists, it should be removed or narrowed, and the Garante's objections addressed.
- Resource the overseers. The FRA stresses giving data protection and human-rights bodies the powers and expertise to monitor use. A judge can only gatekeep what they can see.
The decree's scraping ban and deletion duties are a good model for other member states. Whether it earns that status depends on its application: how rarely emergency routes are used, whether protests stay outside its reach, and whether outside overseers can check the logs. Tight rules with real oversight can let police use a powerful tool while keeping public spaces open to dissent.