Eigma
EN 中 DE
The EU AI Act in Everyday Life 打开 Eigma
目录 Predictive policing

Policing and borders

Predictive policing

The central line is between predicting crime from a person’s profile and assessing evidence connected to actual criminal activity. Since 2 February 2025, Article 5(1)(d) has prohibited providers, deployers and other operators from using AI to predict an individual’s offending risk solely from profiling or personality traits. For law-enforcement systems, the Article 74(8) market-surveillance authority enforces that prohibition.1

The same rule does not prohibit AI supporting a human assessment of someone’s involvement in crime when the assessment is already based on objective, verifiable facts directly linked to criminal activity. The providers and police deployers remain accountable, the Article 74(8) authority enforces the boundary, and the exception has applied since 2 February 2025. Commission guidance says a system using past criminal behaviour and other verifiable facts to help identify a suspect may be high-risk rather than prohibited.2 Adding a person to review a score does not make profiling-only prediction lawful.

From 2 December 2027, Annex III point 6(d) is scheduled to classify other individual offending and reoffending assessments as high-risk. Under Articles 9 to 16, providers will have duties covering risk management, data governance, logging, documentation, oversight design and system performance. Under Article 26, police deployers will have to appoint competent oversight, monitor use, suspend risky operation and retain controllable logs. Article 74(8) authorities will enforce both sets of duties.1 Article 27 will also require public-law deployers, enforced by the same authority from that date, to assess fundamental-rights effects before first use.3 These are duties on organisations, not personal rights to a particular outcome.

Since 2 August 2026, Article 85 has allowed a person who has grounds to suspect an infringement to complain to the relevant market-surveillance authority. Article 86 also provides for an explanation from the deployer after a significantly adverse qualifying decision, subject to legal restrictions and only where Union law does not already provide that explanation right. Its operation before the Annex III rules begin remains unresolved. The Act sets no request format, response deadline or dedicated appeal route, and the Commission’s national contact list was still incomplete on 12 September 2026.14

For police-held personal data, the Law Enforcement Directive may be more useful. Articles 14 and 16 require the police controller to handle access and correction requests, subject to lawful restrictions. Article 11 limits solely automated significant police decisions. National data-protection authorities and courts enforce these rules, which Member States had to implement by 6 May 2018.5

Boundary. The AI Act does not cover systems used exclusively for military, defence or national-security purposes. Police-data access can also be restricted where necessary and proportionate to protect investigations, security or other people’s rights.15

参考文献

  1. [1]Consolidated Artificial Intelligence Act
  2. [2]European Commission guidance on law-enforcement AI
  3. [3]Regulation (EU) 2026/1744
  4. [4]Commission list of AI Act market-surveillance authorities
  5. [5]Law Enforcement Directive
测验

  1. If you suspect that inaccurate police data influenced an AI-assisted assessment, which route may address the underlying data directly?

    • Request police data access and correction under national Law Enforcement Directive rules
    • Send a general complaint only to the European Commission's AI Office instead of requesting correction from the national police
    • Wait for the later high-risk safeguards before challenging the underlying data

    The Law Enforcement Directive can provide access and correction routes now, although lawful restrictions may protect investigations or security.

  2. An offending-risk prediction based only on profiling remains ____ even if a person later sees the score.

    • prohibited in principle
    • permitted as high-risk
    • excluded from review

    Human review after profiling does not supply the objective, verifiable facts required for the permitted evidence-based assessment.

  3. The later high-risk classification turns evidence-based offending assessments into prohibited systems.

    • True
    • False

    High-risk classification brings organisational safeguards; it does not prohibit assessments grounded in objective, verifiable facts linked to criminal activity.

评论

还没有评论，来说第一句吧。