Domain 6 of Annex III AI Act covers AI systems in investigation and law enforcement. The Commission guidelines of 19 May 2026 devote over fifteen pages to it, with sharp delineation against the prohibited practices of Article 5. For police, prosecutors, customs and other enforcement agencies, classification here is particularly delicate.
For the general framework, see the main article on the Article 6(3) filter. For all domains, see the hub overview.
The Five Use Cases of Domain 6
- Point 6(a) Assessing the risk of a person becoming the victim of a criminal offence
- Point 6(b) Polygraphs and similar tools
- Point 6(c) Assessing the reliability of evidence
- Point 6(d) Assessing the risk of (re)offending by concrete persons
- Point 6(e) Profiling of natural persons in detection, investigation or prosecution
Difference with Article 5: The Predictive Policing Prohibition
The guidelines emphasise that Article 5(1)(d) prohibits AI systems that predict the risk that a natural person will commit a criminal offence, solely based on profiling or personality traits. That is therefore prohibited, not high-risk.
High-risk under Point 6(d) concerns recidivism risk assessments containing additional elements, or focused on concrete persons already assessed rather than general predictions.
Prohibited under Article 5:
- AI labelling neighbourhoods or population groups predictively for crime risk
- AI predicting solely from personality profiles whether someone will become a criminal
High-risk under Point 6(d):
- AI estimating recidivism risk of a convicted person during detention or reintegration
- AI assessing flight risk or safety risk in an ongoing investigation
Use Cases in Detail
Point 6(a): Victim Risk
High-risk:
- AI predicting who is at risk of becoming a victim of domestic violence
- AI flagging vulnerable persons for preventive intervention
Point 6(b): Polygraphs and Similar Tools
High-risk:
- AI in modern lie detection (stress, micro-expressions, voice)
- AI doing "credibility assessment" during interrogations
Point 6(c): Evidence Assessment
High-risk:
- AI assessing authenticity of digital evidence
- AI analysing consistency of witness statements
Filter possible:
- AI only indexing or categorising evidence material without assessment
- AI generating transcripts of interrogation videos
Point 6(d): (Re)offending Risk
High-risk:
- Probation tools scoring recidivism risk
- Risk assessment in pre-trial detention decisions
Point 6(e): Profiling During Investigation
High-risk:
- AI profiling suspects based on data from ongoing investigation
- AI generating "person of interest" lists from data analysis
Outside scope:
- Forensic analysis of a specific crime scene (DNA, fingerprints)
- Pure object detection in imagery without person profiling
Sector-Specific Pitfalls
Pitfall 1: Check Article 5 First
For policing applications, Article 5 prohibition often looms. Always start classification with Article 5, not Annex III. A prohibited system does not "still" become high-risk.
Pitfall 2: Prosecution Tools
For prosecution agencies, many "case management" tools can fall under Point 6(e) once they analyse personal data for investigative purposes.
Pitfall 3: Sector-Specific Laws Still Apply
Many enforcement tools connect to national registers, Schengen Information System, EURODAC. The AI Act sits on top of existing police data legislation and sectoral safeguards, not instead.
What to Do
Split per use case
Risk assessment of persons (6(a), 6(d)), polygraph-like (6(b)), evidence assessment (6(c)), profiling (6(e)). Each has its own examples and filter possibilities.
Align with police law
The AI Act is a layer on top. Police data legislation and specific laws on investigative powers continue to apply.
Secure human oversight
For high-risk AI in policing, human oversight under Article 14 is not optional. Decisions with legal consequences cannot be solely AI-based.
Test your AI against the Article 6(3) filter
Free interactive self-assessment, updated for the Commission guidelines of 19 May 2026. 9 steps, personal report with reasoning, vendor questions and next steps.