AI, Algorithms, and Automation in Government: Investigating the ethical, legal, and practical implications of using artificial intelligence for decision-making in social services, policing, and welfare distribution

This systematic review of 87 studies (2015–2025) examines how government AI affects social services, policing, and welfare distribution. Three persistent ethical challenges emerged: algorithms produce biased outcomes that harm disadvantaged groups, automated systems lack transparency for contesting decisions, and machine learning produces uncontrollable decisions with no identifiable responsible party. Legal frameworks in the US, EU, Canada, and Australia remain incomplete. US courts face trade secret protections that block discovery of evidence; the EU’s GDPR allows human evaluation to bypass these protections through “solely automated” exceptions. No jurisdiction has established special courts for algorithmic appeals. Three implementation obstacles cause permanent system damage: biased training data, feedback loops, and digital divide exclusions. Citizens experience severe outcomes, including service avoidance and extreme psychological harm. In the Dutch childcare case, 87% of affected families reported anxiety, 63% depression, and 12% suicidal ideation, alongside material deprivation, family separation, and lost trust in public institutions. Policing has the most developed legal framework, yet operational bias persists, while welfare distribution offers the weakest legal protections, directly impacting the most disadvantaged groups. Government AI systems require structural technical, organisational, legal, and political changes to prevent unprecedented automated discrimination. This review proposes a temporary halt on dangerous systems while establishing requirements for third party audits, meaningful human assessment, algorithm based appeal procedures, and community based design methods. It contributes the first cross domain comparison of AI governance across social services, policing, and welfare, identifies “token human review” as a systematic bypass of legal protections, and documents psychological harm rates (87% anxiety, 12% suicidal ideation) requiring new remediation frameworks.