The point: Germany plans to allow tax authorities to use real taxpayer data to train AI models, which conflicts with EU data protection principles but is intended to be limited by deletion deadlines and control obligations.
The Federal Finance Ministry has submitted a draft of the 2026 Annual Tax Act that would give tax authorities the right for the first time to train artificial intelligence using unredacted, real tax data from citizens. This represents a significant departure from previous data protection principles and could reignite debates across the EU about AI use in government agencies.
The draft proposes an amendment to Section 29c of the German Tax Code that would allow tax offices to use personal tax data for the development, optimization, and testing of automated procedures. The goal is to accelerate batch processing of tax returns and reduce administrative burden – a concern that is increasingly coming to the fore given the shortage of skilled personnel in tax administration.
This initiative conflicts directly with the General Data Protection Regulation (GDPR). Previously, the principle of purpose limitation prevented authorities from using tax data for AI training – personal data may only be used for its original collection purpose, in this case tax assessment. However, the Federal Finance Ministry argues that synthetic or fictional test data is insufficient for realistic training: real case constellations are necessary to map complex data structures and cross-agency information linkages.
To address data protection concerns, the draft includes specific technical and regulatory safeguards. All personal data used for training must be irreversibly deleted no later than one year after the end of the training measure. Additionally, the AI is restricted to a supporting role: it is to perform preliminary reviews and alert case officers to anomalies; final decision-making authority remains with humans. Purely automated decisions adverse to taxpayers are not permitted.
Germany is thus following an international trend: Spain’s tax authority already uses automated systems to forecast error rates and detect tax evasion early, and Belgium deploys AI for real-time monitoring of VAT transactions. However, the German approach is being viewed critically in data protection circles, as the materiality of the processed data – millions of tax returns containing sensitive information – places considerable strain on the envisaged safeguards.
Source: www.it-daily.net · Published 2 June 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification through Lumi News Pipeline v1.2.9.