Estonia Builds AI Tool to Catch Legislative Errors After €24M Tax Blunder

A Costly Typo Sparks an AI Solution

A single phrase in Estonia's revised Gambling Tax Act cost the government €24 million ($27.4 million) in lost annual revenue. When parliament passed changes in December intended to lower remote gambling tax rates, the law's wording mistakenly referenced only "skill games" for that year—accidentally exempting online casinos from taxation entirely.

The error was first caught by legal counsel for a gambling operator. But when Luukas Ilves, Estonia's former undersecretary for digital transformation, ran the legislation through Claude and Gemini, both AI systems immediately flagged the inconsistency. Within hours, Ilves built a prototype called Apsakaleidja—"Fuckup Finder"—that pulls draft bills from parliament's website and identifies broken references, contradictory wording, arithmetic errors, and impossible dates, according to WIRED, which first reported the details.

The tool currently lists 112 bills, rating 102 as high risk. Ilves demonstrated it on national television, where it highlighted contradictory wording in draft legislation.

Why it matters

Estonia's response reveals how governments might use AI not just to catch mistakes after the fact, but to fundamentally reshape how they draft and review policy. For a country whose gambling industry is worth approximately €300 million ($343 million) and hosts one of the EU's fastest-growing online gambling markets, the stakes are substantial. The incident is accelerating Estonia's already ambitious digital governance agenda—and raising questions about where human judgment remains essential.

From Embarrassment to National Strategy

Prime Minister Kristen Michal told WIRED the blunder "demonstrated that AI can be an incredibly useful assistant" and showed how such tools can empower citizens. In January, he launched Eesti.ai, a program aimed at doubling national productivity by 2035 through AI adoption. Advisers include Bolt founder Markus Villig and Ilves.

In April, parliament received legislation granting state and local government authority to use AI for automating administrative processes. That bill is currently under debate. In June, Michal announced plans for Estonia to become the first country creating official digital identities for AI agents.

Drawing Lines Around Automation

Kirke Maar, team lead of Eesti.ai, outlined Estonia's framework for AI decision-making. "Rule-bound decisions" based on verifiable facts—where meeting criteria automatically determines outcomes—are considered appropriate for automation. Tax declarations, already prefilled in Estonia, could evolve into AI agents preparing complex filings end-to-end, with citizens confirming or intervening as needed.

But "a decision requires genuinely weighing competing interests or judgment about a person's specific circumstances, a human belongs in the loop from the start," Maar said. Citizens can invoke their right to be heard at any point, immediately ending automated procedures. Every automated decision must leave an audit trail showing what data was used, which rules applied, when decisions were made, and how citizens can challenge them.

Catherine Flick, who researches technology ethics at the University of Staffordshire, questioned why human review processes failed to catch the gambling tax error initially. "At some point someone has to sit down and read through the whole thing, with the understanding of the context," she told WIRED.

Michal emphasized that AI serves as assistant, not authority. "AI does not replace democratic institutions, the constitution, or the will of voters," he said. "If AI identifies a mistake in legislation, it is no different from a human spotting one. The responsibility to correct it remains with parliament, the courts, or the public administration."

The details were first reported by WIRED.