How a $28 Million Tax Blunder Sparked Estonia's 'Fuckup Finder' AI Tool

A single phrase in Estonia’s gambling tax law cost the government an estimated €24 million ($27.4 million) in lost revenue last year. The mistake? The December amendment to the Gambling Tax Act only mentioned “skill games” for that year, accidentally leaving online casinos—a fast-growing sector worth around €300 million—completely outside the tax net. A legal counsel for a gambling operator caught the error, but the real surprise came when former digital undersecretary Luukas Ilves ran the legislation through AI chatbots Claude and Gemini. Both systems flagged the inconsistency immediately.

Within hours, Ilves built a prototype called Apsakaleidja—Estonian for “Fuckup Finder.” The tool pulls draft bills from parliament’s website and highlights broken references, contradictory wording, arithmetic errors, and impossible dates. Out of 112 bills currently listed, 102 are rated high risk. Ilves even demonstrated it live on national TV, leaving the host stunned. Prime Minister Kristen Michal told WIRED the incident proved AI can be an incredibly useful assistant, and it sparked a broader push: the Eesti.ai program aims to double productivity by 2035 through AI upskilling, with advisors including Bolt founder Markus Villig and Ilves himself.

In April, the government introduced a bill allowing state and local agencies to use AI for administrative automation, now moving through parliament. By June, Michal suggested Estonia could become the first country to create official digital identities for AI agents. The idea is that for rule-bound decisions—like tax filings or benefit eligibility—automation can work without human intervention, with citizens simply notified of outcomes. For discretionary choices involving competing interests, a human stays in the loop from the start. Critics like ethics researcher Catherine Flick ask why humans aren’t already doing this review, but Estonia’s digital infrastructure—99 percent of public services online, a decade of digital identity leadership—makes it uniquely positioned. The government insists AI is an assistant, not an authority, and every automated decision must leave an audit trail, with citizens able to challenge or escalate to a human official at any point.