Governments across the world are using automated risk profiling systems to identify potential offenders before any offence has been committed, and those systems are incompatible with international human rights law. That is the central finding of a new report by Amnesty International, the first comprehensive assessment of risk profiling against international human rights standards.

Risk profiling, the algorithmic prediction of whether a person or group is likely to break a law or rule, has been deployed in law enforcement, social security and migration control across multiple jurisdictions, including Denmark’s social benefits fraud investigations, France’s residence permit system, the Netherlands’ social fraud detection programme and Australia’s automated debt recovery scheme. Governments justify these tools as cost-effective, objective and crime-reducing. Amnesty disputes all three claims.

The scientific foundations are weak, the report argues. Predicting whether an individual will commit fraud or a criminal offence requires data of a quality and completeness that does not exist and never will. Proxies are invariably used instead: re-arrest rates stand in for re-offending; administrative errors substitute for fraudulent social security claims. These proxies embed and amplify existing biases. Because marginalized communities are already over-policed and under-served, their historical data makes them appear statistically riskier. The algorithm then acts on that appearance, reinforcing the original discrimination in a self-perpetuating cycle.

The human costs are serious. People targeted by risk profiling in high-stakes contexts face psychological distress, stigmatisation, wrongful denial of benefits, deportation and imprisonment. The systems disproportionately flag racialized people, Muslims, people with disabilities, those on low incomes and people on the move as suspicious. Transparency is typically absent, leaving those affected unable to challenge decisions that may destroy their livelihoods or their right to remain.

Amnesty’s recommendation is unambiguous: risk profiling in law enforcement, migration and social security must be banned outright, whether the systems are driven by artificial intelligence or rule-based, and whether or not a human formally takes the final decision. Until such bans are legislated, authorities should halt both the use and the development of these tools. The veneer of objectivity, the report concludes, is precisely what makes them dangerous.

Sources: Amnesty International, “Automating Suspicion: Risk Profiling as a Smokescreen for Structural Discrimination and Inequality,” June 10th, 2026; statement by Alexander Laufer, Amnesty International Netherlands Researcher on Technology and Human Rights, June 2026.