Predictive policing covers algorithmic systems "designed to 'predict' behavioral patterns and traits," which "analyze video feeds, location history, communications metadata, and other information to assign risk scores or forecast future actions" and "serve to justify the intervention of law enforcement authorities and security forces to 'anticipate' illicit behaviors" (No Fair Play: Mapping the 2026 World Cup Surveillance Stack (Ranking Digital Rights, July 2026)).
The stated harms
Two are commonly identified. The first is amplification: these systems "are trained with historically biased datasets and risk amplifying bias," so a model learns from enforcement records that reflect where police have previously concentrated rather than where offences have occurred, and directs further attention there. The second concerns process rather than accuracy — Ranking Digital Rights argues that "data extraction schemes and 'likelihood scores' violate the right to a fair trial and privacy," because a score generated before any offence attaches consequences to a prediction that the subject cannot contest.
Position in the surveillance stack
Ranking Digital Rights places predictive policing as one of five layers in its mapping of the 2026 World Cup deployment, between data harvesting at scale and communications interception. The layering matters analytically: predictive systems consume the output of the harvesting layer and feed the data-integration and command-centre layer, so their inputs and their consequences are both determined by adjacent systems rather than by the predictive model itself.
The integration layer is where the report locates the qualitative change, since merging feeds "eliminat[es] the fragmented safeguards that existed while the systems remained separate" and "may also create permanence: data that might have been deleted in individual systems persists indefinitely in integrated databases." A risk score derived from one jurisdiction's data can therefore follow a person into another's.
US federal deployment
Reporting by Joseph Cox for 404 Media, public by September 8, 2026, describes US Border Patrol operating predictive policing units that analyse Americans' financial activity and other data and pass the resulting intelligence to local police, who then stop people not suspected of any specific crime. The Marshall Project indexes the units under the name Predictive Intelligence Targeting Teams (Source: 404media.co; themarshallproject.org). The 404 Media article body sits behind a paywall and only its opening paragraph was retrieved, so the account here is limited to that summary; the unit's inputs, models, scale and legal authority are not established.
The described pattern differs from the municipal deployments the surveillance-stack framing covers in two respects: the predicting agency and the intervening agency are different bodies, and the input is financial-activity data rather than enforcement records or video. Both bear on the process objection above, since a person stopped on a federal score has no visible decision to contest. See CBP — Customs and Border Protection (AI Deployer).
Regulatory position
The EU AI Act prohibits individual predictive-policing systems that assess the risk of a person committing a criminal offence based solely on profiling or personality traits, among its prohibited practices. US regulation is largely municipal, which is what allows event-driven procurement to expand deployment without a general legislative decision — the mechanism Ranking Digital Rights traces through South Africa 2010, Brazil 2014–16, Qatar 2022 and Paris 2024.
The United Kingdom has no equivalent prohibition. Parliament's Joint Committee on Human Rights recorded in September 2026 that AI is used in the UK public sector for predictive policing, among other purposes, and cited evidence that a Durham Constabulary system for assessing the risk of recidivism placed heavy reliance on individuals' postcodes and locations, potentially giving those factors a significant effect on prosecution decisions, as well as tools used in prisons for security categorisation and violence-risk assessment that witnesses said put minoritised groups most at risk. Amnesty International UK's Javier Ruiz Diaz said such risk-profiling systems do not say "we think this is happening"; they say "we know this is happening." The committee recommended that a new AI Bill prohibit uses of AI inconsistent with human rights, including "inappropriate use of profiling," with the detail set after public consultation (Recommendations 45–46) (Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026)).
Relationships
- instance-of: AI and Surveillance
- related: Surveillance Technology, AI Bias and Discrimination, No Fair Play: Mapping the 2026 World Cup Surveillance Stack (Ranking Digital Rights, July 2026), EU AI Act (Regulation 2024/1689), ICE — Immigration and Customs Enforcement (AI Deployer)
- related: Human Rights and the Regulation of AI (Joint Committee on Human Rights, HC 160, September 2026) — UK public-sector use and a recommended statutory prohibition on inappropriate profiling