Tax Office 2.0: Tax authorities to train AI with real citizen data

The digitalization of the German tax administration is set to accelerate, which should make data protectionists and IT experts prick up their ears. With the draft law from the Federal Ministry of Finance for the Annual Tax Act 2026, the government is preparing for the widespread introduction of artificial intelligence systems in tax offices. According to its statement, the legislator is reacting to the need for action and the necessity to efficiently manage complex German tax law in the digital age.

According to the draft, the technological upgrade is primarily intended to help reduce the administrative burden in the authorities and speed up the processing of tax returns in mass procedures. For taxpayers, this means, in return, that their sensitive financial data will be used in the future to optimize state algorithms.

The core of the new regulation is hidden behind an inconspicuous amendment to the Tax Code (Abgabenordnung). Specifically, it concerns Section 29c, which regulates the use of personal data for data processing purposes. Previously, it was very difficult for tax authorities to feed AI systems with real tax data due to the purpose limitation principle of the General Data Protection Regulation (GDPR).

Now, the legal permission is to be granted to use real, unaltered tax data for the development, testing, and modification of automated procedures. Training with fictitious test data is not effective, the reasoning states, as a modern AI system loses its functionality or delivers inaccurate results without access to real data structures and contexts. Especially when information from different administrative systems needs to be clearly linked, the creation of artificial test cases reaches economic and logical limits.

Real tax data for accurate algorithms

To nevertheless comply with data protection requirements, the legislator is to link AI training with a protective measure: the personal training data used must be irrevocably deleted at the latest one year after the end of the respective development or testing measure, according to the plan. This deadline is intended to prevent permanent, transparent profiles of taxpayers from remaining in the systems for research purposes.

Furthermore, the finance department assures that AI in tax proceedings will only function as an auxiliary tool. The decision-making authority and the final review will remain with the human case worker in the tax office. There will be no automated, purely algorithmic tax assessments that are disadvantageous to citizens. The AI is primarily intended to act as a preliminary reviewer that detects irregularities.

To increase attractiveness for research companies, the draft provides for an expansion of tax research funding. Companies will be able to claim up to 25 million euros per year in tax relief for investments in research and development in the future. Previously, the cap was 15 million euros. In this way, the ministry wants to set a strong tax incentive to retain and expand private investments in future technologies such as AI, semiconductor technology, and green energy domestically.

(nen)