Law Enforcement AI: High-Risk or Prohibited?

Both — depending on what the system does. AI used in policing to assess a person's risk of offending or re-offending, to act as a polygraph, to evaluate the reliability of evidence, or to profile people during an investigation is high-risk under Annex III point 6 of the EU AI Act, with obligations from 2 December 2027. But two adjacent law-enforcement uses are prohibited outright under Article 5: predicting offending solely from profiling (5(1)(d)), and real-time remote biometric identification in public spaces (5(1)(h)). Getting on the wrong side of that line is the most expensive mistake in this domain.

This page draws the prohibited-versus-high-risk boundary precisely, then sets out the FRIA and the non-public EU-database rule that apply to the high-risk uses.

---

What Annex III point 6 covers

Annex III point 6 lists law-enforcement AI uses that are high-risk — lawful, but subject to the full regime. They include AI intended to be used by or on behalf of law-enforcement authorities, or by EU bodies supporting them:

• to assess the risk of a natural person offending or re-offending, where that assessment is not based solely on profiling (the part that is solely-profiling-based is prohibited, below);
• to assess the risk of a person becoming the victim of a criminal offence;
• as a polygraph or similar tool;
• to evaluate the reliability of evidence in the course of investigation or prosecution;
• to profile natural persons in the course of detection, investigation or prosecution of criminal offences.

These are high-risk because the stakes for liberty and due process are acute, but they support a human-led criminal-justice process rather than substituting an automated verdict. They carry the provider stack and the deployer duties — not a ban.

---

The prohibited line: where point 6 stops

Two prohibitions in Article 5 sit right next to point 6. Cross into them and there is no compliance route — the use is banned, and breaches carry the top penalty of up to €35 million or 7% of total worldwide annual turnover (Article 99(3)). Both bans have been in force since 2 February 2025.

• Article 5(1)(d) — predicting offending solely from profiling. AI that makes risk assessments of natural persons to assess or predict the likelihood of committing a criminal offence, based solely on profiling or on assessing personality traits and characteristics, is prohibited. The high-risk version in point 6 is the one that supports a human assessment grounded in objective, verifiable facts directly linked to a criminal activity. The fulcrum is the word "solely": profiling-only prediction is banned; fact-grounded risk assessment that informs a human decision is high-risk.
• Article 5(1)(h) — real-time remote biometric identification. Live facial recognition in publicly accessible spaces for law-enforcement purposes is prohibited, save for narrowly defined, pre-authorised exceptions (e.g. a targeted search for a specific victim or a serious imminent threat). Post (non-real-time) biometric identification is not banned — it is Annex III point 1 high-risk.

---

FRIA, human oversight, and the non-public EU database

For the high-risk point 6 uses, the deployer is almost always a public authority — so the Article 27 Fundamental Rights Impact Assessment is mandatory before first use, alongside the Article 26 deployer duties and Article 14 human oversight. A re-offending score may inform a decision; it may not make it. The provider carries the Article 16 stack, including the Annex IV technical documentation (Article 11) and conformity assessment (Article 43).

Registration has a special twist here. High-risk systems must be registered in the EU database (Articles 49 and 71) — but for high-risk AI used in law enforcement, migration, asylum and border control, registration goes into a non-public (secure, restricted) section of that database, not the public-facing register. The duty to register still applies; the entry is simply not published. Don't tell a police authority it is exempt from registration — it isn't; the visibility is just restricted.

---

Worked example: a regional Polizeipräsidium

A regional police headquarters, the Polizeipräsidium, evaluates three tools. (1) A re-offending risk assessment used on individuals already charged, drawing on objective case facts and feeding a documented human decision — high-risk under Annex III point 6: lawful, but it needs Article 14 human oversight, a mandatory Article 27 FRIA, the provider stack, and registration in the non-public section of the EU database. (2) A vendor pitch for a tool that scores members of the public on their "likelihood to offend" from demographic and behavioural profiling — this is prohibited under Article 5(1)(d); there is no documentation route and it cannot be deployed. (3) A proposal to run live facial recognition across the city's CCTV to flag passers-by in real time — prohibited under Article 5(1)(h) outside the narrow authorised exceptions. Two of the three are banned; only the fact-grounded risk tool can proceed, and only with the full high-risk apparatus.

---

How Confir helps

Confir tests a law-enforcement use against the Article 5 prohibitions before Annex III, so a solely-profiling crime-prediction tool or a real-time public biometric system is flagged as prohibited rather than scoped as "high-risk to document". For genuine point 6 uses it assembles the provider stack, flags the Article 27 FRIA as mandatory, and notes that registration goes into the restricted section of the EU database. The engine is deterministic and rule-based, each finding cited to the Article or Annex point that fired.

Related guides

• Law enforcement risk AI under Annex III
• Crime prediction and the prohibited line
• AI polygraph systems
• Evidence evaluation AI
• Real-time face recognition in public is prohibited
• Article 5 prohibited practices