Law Enforcement , Annex III Point 6 High-Risk AI
AI used by or on behalf of law enforcement authorities, or by Union institutions supporting them, for risk assessment of natural persons, polygraph-style deception detection, evidence evaluation, offence prediction, or profiling in investigations is high-risk under Annex III Point 6.
Scope, Definitions and Boundary with Prohibited Practices
Annex III Point 6 covers the law-enforcement domain narrowly defined: AI used by or on behalf of law enforcement authorities, or by EU institutions supporting them. The sub-points map to: individual risk assessment, polygraph-style systems, evidence reliability evaluation, offence prediction and recidivism, profiling of natural persons, and crime analytics with prediction dimensions.
The Article 5 prohibitions bite hard here. Predictive policing based solely on profiling or assessed personality traits is prohibited under Article 5(1)(d), except where it supports human assessment of criminal involvement based on objective, verifiable facts directly linked to a criminal activity. Untargeted facial-image scraping for database building is prohibited under Article 5(1)(e). Emotion recognition is prohibited in workplaces but permitted as high-risk in investigative contexts outside those settings.
Example Systems Under Annex III Point 6
Illustrative examples, each Member State's market surveillance authority may refine classification guidance over time.
Example
Individual recidivism risk scoring for bail decisions
Point 6(a) high-risk. Article 5(1)(d) prohibition must be assessed, if 'solely' profiling-based, prohibited.
Example
AI-assisted CSAM detection for Europol
Point 6(c) / 6(d) high-risk, evidence reliability and offence-prediction features.
Example
Real-time facial recognition at a protest (law enforcement)
Article 5(1)(h) prohibited except narrow exceptions. If used retrospectively on recorded footage, falls within Point 1 / Point 6 high-risk scope.
Example
Analytics tool that searches open-source for suspect identification
Generally high-risk under Point 6(e). Data source legality (GDPR) and Article 5(1)(e) scraping prohibition must both be evaluated.
Example
Administrative investigations by non-law-enforcement authorities
Point 6 applies specifically to law enforcement, other authorities may fall under Point 5 or Point 7.
The Articles That Apply
High-risk classification under Annex III triggers the full Chapter III, Section 2 obligation set. Deployers pick up additional obligations under Chapter III, Section 4.
| Article | Obligation | What It Means in Practice |
|---|---|---|
| Art. 9 | Risk Management System | Law-enforcement AI carries the highest risk of fundamental-rights harm and requires correspondingly rigorous risk management, including specific consideration of procedural rights and the presumption of innocence. |
| Art. 10 | Data Governance | Training data from policing sources carries durable historical bias. Article 10(5) permits processing of special category data for bias detection under strict conditions, the conditions are stricter for law enforcement. |
| Art. 14 | Human Oversight | Natural persons taking decisions must have competence, resources, time and authority. Over-reliance on AI output in charging or bail decisions is a well-documented failure mode. |
| Art. 26 | Deployer Obligations | Member State law-enforcement authorities are the primary deployers. Extra-territorial application to non-EU providers supplying them is broad. |
| Art. 27 | FRIA, MANDATORY | Mandatory for law-enforcement deployers (public authority). Must be submitted to the market surveillance authority. |
| Art. 49 | EU Database Registration | Law-enforcement Annex III systems register in a non-public section of the EU database, a partial exception from the full publication requirement. |
| LED (2016/680) | Law Enforcement Directive | Parallel regime on processing of personal data for law-enforcement purposes. Stacks with AI Act obligations. |
| CFR | Charter of Fundamental Rights | Particular attention to Articles 6 (liberty), 7 (private life), 8 (data protection), 47 (fair trial) and 48 (presumption of innocence) of the Charter. |
Pitfalls to Avoid on Annex III Point 6
Recurrent patterns seen in pre-enforcement readiness assessments.
Predictive policing based 'solely' on profiling
Article 5(1)(d) is drafted as a narrow prohibition but is interpreted strictly: if the AI output materially drives the policing decision without independently verifiable fact input, it is prohibited.
Facial image scraping from social media for database building
Article 5(1)(e) prohibits untargeted scraping of internet or CCTV sources to create or expand facial recognition databases. This is a categorical prohibition, no law-enforcement carve-out.
Relying on LED/GDPR compliance as AI Act compliance
The Law Enforcement Directive regulates processing; the AI Act regulates the behaviour of the AI system. Two regimes, two sets of obligations.
Cross-border data flows from non-EU providers
Non-EU providers supplying law enforcement must appoint an authorised representative (Article 22) and meet extra-territorial scope tests in Article 2(1)(c). The bar is higher, not lower, for this domain.
Operationalise Point 6 Compliance
The Full Readiness Bundle gives your legal and compliance teams the 58-Point Compliance Checklist, Annex III Classification Matrix, FRIA template, Annex IV Technical Documentation Checklist, and 7 more documents, all aligned to the Annex III high-risk regime (now applying from 2 December 2027 under the May 2026 Omnibus deal).
one-time · instant download
Get the Full Bundle →Also available: Checklist Pack $149 · White-Label $999/yr
Next Steps
Deadline Planning
15-week compliance work plan for the EU AI Act's phased timeline.
Week-by-Week Plan →