Estonia’s AI tool to flag legal errors in draft legislation after $28 million tax blunder

A single wording error in a December 2025 amendment to Estonia’s Gambling Tax Act resulted in online casinos being excluded from the tax net for a year, costing the government €24 million ($27.4 million) annually in lost revenue. The mistake was identified by a legal counsel for a gambling operator, but it was also flagged within hours by AI systems, including Claude and Gemini, when Luukas Ilves, former undersecretary for digital transformation, tested the legislation using the models.

Within hours of discovering the error, Ilves built a prototype AI tool called Apsakaleidja—Estonian for “Fuckup Finder”—to scan draft bills from the Riigikogu website for inconsistencies such as broken references, contradictory wording, arithmetic errors, and impossible dates. The tool categorizes flagged issues by risk level, with 102 of the 112 currently listed draft bills rated as high risk.

The incident prompted a broader government push to integrate AI into legislative review and public administration. In January 2026, Prime Minister Kristen Michal announced plans to use tools like Apsakaleidja not only to review draft legislation but also to draft new laws, aiming to preemptively identify and fix loopholes. The government also launched Eesti.ai, an initiative to upskill Estonians in AI use with the goal of doubling national productivity by 2035.

The Eesti.ai program includes advisers such as Markus Villig, founder of Bolt, and Ilves, the creator of Apsakaleidja. In April 2026, the government submitted a bill to parliament that would grant state and local governments the right to use digital solutions, including AI, to automate administrative processes. The bill is currently under debate.

By June 2026, Michal stated that Estonia aims to become the first country in the world to create official digital identities for AI agents, emphasizing the need for agility and adaptability in the public sector as AI technologies evolve. Estonia’s digital-first governance model, which includes 99 percent of public services being online, provides a foundation for this expansion.

The government is debating how to delineate the roles of AI and human oversight in administrative decisions. Kirke Maar, team lead of Eesti.ai, describes a framework where automation is appropriate for rule-bound decisions—such as tax declarations, which are already prefilled in Estonia—while human involvement remains essential for discretionary decisions that require judgment or the weighing of competing interests.

Every automated administrative decision would include an audit trail detailing the data used, the rule applied, the timing of the decision, and the process for challenging or correcting it. Additionally, citizens would retain the right to invoke their right to be heard, at which point the automated procedure would halt and a human official would take over. Automated decisions would also be paused if a citizen disputes the outcome.

Accountability is a central focus of Estonia’s AI strategy. Liina Vahtras, managing director of Estonia’s e-residency program, emphasizes the importance of clear chains of responsibility as AI agents interact with public services, banks, and other digital systems. The government aims to ensure that the authorization, scope, and responsibility for AI actions are transparent and traceable.

Prime Minister Michal has underscored that AI is intended to assist, not replace, democratic institutions. He stated that while AI can identify errors in legislation, the responsibility to correct them remains with parliament, the courts, or public administration, reaffirming that AI does not supplant constitutional or voter-driven decision-making.