Accountability for Artificial Intelligence Errors in Public Administration: Legal Analysis and Lessons from Real-World Cases

Governments increasingly use artificial intelligence (AI) to make or support decisions that directly affect citizens, such as detecting fraud, assigning benefits, and processing applications. When such systems make mistakes, a basic democratic question arises: who answers to the people? This project examines accountability for AI errors in public administration. It begins by describing the Law of the Republic of Kazakhstan "On Artificial Intelligence" (No. 230-VIII, signed on 17 November 2025 and in force since 18 January 2026), its adoption, and its provisions on liability, with a brief comparison to the European Union's AI Act. It then argues that AI itself cannot bear responsibility, since it is not a legal subject, and evaluates which human and institutional actors (developers, operators, officials, the state) should be held accountable. Finally, it tests these arguments against two real cases of automated decision-making failures, in the Netherlands and Australia, using a single set of comparative criteria. The study aims to identify possible gaps in accountability and to propose recommendations for Kazakhstan.