How Does the AI Act Apply to Law Enforcement AI Systems?

Which law enforcement functions are listed?

Annex III point 6 lists eight kinds of law enforcement use, each of which must be tested on its own terms. They include individual risk assessments of the likelihood of offending or reoffending, polygraphs and similar tools to detect the emotional state of a person, systems to detect deepfakes, systems to evaluate the reliability of evidence in criminal proceedings, systems to predict the occurrence or reoccurrence of an actual or potential criminal offence based on profiling or on personality traits and past criminal behaviour, systems for profiling natural persons in the detection, investigation or prosecution of criminal offences, and crime analytics systems for patterns relating to criminal activity. A system that performs any of those functions is a candidate for high-risk classification, and a system that performs more than one must be assessed for each. The record should describe the function, the data, the output and the decision it influences, citing the Annex point for each function, rather than describing the system only as a law enforcement tool.

What must be screened under Article 5?

Law enforcement systems must be screened under Article 5 before classification, with particular attention to biometric identification, biometric categorisation, emotion inference, social scoring and predictive policing prohibitions. Real-time remote biometric identification in publicly accessible spaces for law enforcement is prohibited subject to defined exceptions and safeguards, and must be tested element by element. Systems that evaluate or classify persons based on social behaviour or personal characteristics, to infer a social score leading to detrimental treatment, can engage the social scoring prohibition. Systems that assess the risk of offending based on profiling or personality traits must be tested against the predictive policing provisions in Article 5 and Annex III together. The screen should examine the input data, the inferences, the population and the decisions that follow, and should record the conclusion for each prohibition. A law enforcement function that is prohibited must not proceed to conformity work, and a function close to a prohibition needs qualified advice before deployment.

For the broader regulatory framework, see finance, insurance and critical infrastructure.

How does restricted registration work here?

Registration for high-risk systems in Annex III points 1, 6 and 7 in the areas of law enforcement, migration, asylum and border control management sits in a secure non-public section of the EU database under Article 49(4), including only specified information from Annex VIII and Annex IX, with access limited to the Commission and the national authorities referred to in Article 74(8). A law enforcement provider or deployer must therefore identify the Annex point and area, enter only the specified items, and use the secure section, rather than entering a fuller record publicly or withholding registration. Deployer registration under Article 49(3) applies to public authority deployers, and the deployer's Section C information is likewise limited to the specified points in the restricted section. The restricted treatment does not remove the duty to document, to assess fundamental rights impact where required, or to keep logs and oversight records: it limits who can see the registration entry, not the underlying compliance work.

Who supervises law enforcement systems?

Supervision of law enforcement systems involves the market surveillance authority and, in the defined areas, the national data protection authority, which acts as the market surveillance authority or works in close cooperation with it, as the consolidated text provides. That allocation reflects the processing of personal data, including biometric and criminal justice data, and the fundamental rights at stake. A law enforcement deployer should therefore coordinate its AI Act and data protection records, assessments and authority contacts, and should expect supervision that tests both regimes together. The deployer should also plan for the fundamental-rights impact assessment where Article 27 requires it, for the information and redress routes available to affected people, and for judicial oversight of evidence and decisions. A file that satisfies the AI Act classification but cannot answer a data protection or fundamental rights question will not withstand supervision in this area, and the file should be built to answer both from the start.

What oversight is needed for consequential decisions?

Oversight for a law enforcement system means a competent person, with time, information and authority, reviews the system's output before a consequential action is taken, considers information beyond the output, and can decide differently from the system. A risk score, a match, a flag or a prediction is an output, not a finding, and it should never be treated as proof of guilt, risk or identity on its own. The overseer should understand the system's limitations, its error rates and its potential for bias, and should record the decision, the information considered and the outcome. The deployer should monitor outcomes, including whether the system performs consistently across groups, whether its outputs are followed or overridden, and whether challenges reveal patterns that require action. Affected people should have a route to challenge a decision and to seek a remedy, subject to the procedural law that governs the investigation or proceedings concerned. An oversight arrangement that confirms the system's output without assessment is not oversight, and the record should be able to show that assessment took place.

Continue with migration, asylum and border control AI.

What does this guide not decide?

This guide does not decide whether any law enforcement system is prohibited, high-risk or authorised, does not determine whether any investigation, prosecution or evidence was lawful, and does not give operational, criminal justice or data protection advice. Those questions depend on the system's function, the Article 5 exceptions and safeguards, the Annex III listing, national criminal procedure and data protection law, applied to the specific facts with qualified advice. The guide is jurisdiction-labelled to the EU AI Act and does not describe the law of the United Kingdom, the United States or any other jurisdiction. Where a law enforcement system is to be deployed, confirm the prohibition, classification, registration, supervision and procedural position with the competent authorities and legal advisers before use, and keep the assessment with the deployment record.

Frequently asked questions

Continue through this AI Act cluster

• finance, insurance and critical infrastructure
• migration, asylum and border control AI
• biometric identification systems
• what is public and what is restricted in the EU database
• prohibited practices under Article 5
• high-risk AI systems