Germany Proposes Using Real Taxpayer Data to Train AI Systems
Germany's Federal Ministry of Finance is seeking legal permission to train artificial intelligence systems on unaltered citizens' financial data-a move that would circumvent current European privacy rules. The proposal, included in the draft Annual Tax Act 2026, aims to deploy AI across the country's tax offices to process returns faster and manage complex tax law.
The change would require amending Section 29c of the German Tax Code to allow the use of real, unaltered tax data for developing and testing automated systems. Currently, the General Data Protection Regulation (GDPR) prevents such use through its purpose limitation principle, which restricts data to its original stated purpose.
Why Real Data Instead of Fictional Data?
The Ministry argues that training AI on fictitious data produces inaccurate results. Real taxpayer information, the reasoning goes, better reflects the complexity and edge cases that tax systems encounter in practice.
The proposal includes a safeguard: personal training data must be permanently deleted no later than one year after training ends. The Ministry also says AI will function only as an auxiliary tool, with humans retaining final decision-making authority.
Privacy and Bias Risks
The International Bureau of Fiscal Documentation, a non-profit foundation, has flagged significant concerns about tax authorities' use of AI. Tax agencies handle large volumes of sensitive personal and financial data-information that can be leaked or misused if security fails.
AI systems trained on incomplete or flawed data can also introduce bias, leading to unfair treatment of taxpayers. The foundation warns these risks apply across European tax systems, not just Germany's proposal.
AI Already in Use Across Europe
Germany would not be alone in deploying AI for tax work. Spain's Tax Agency uses automated systems to flag filing errors and detect non-compliance. Belgium monitors and blocks suspicious VAT transactions. France's tax authority uses machine learning and aerial imagery to identify undeclared swimming pools and property extensions.
These deployments show both the appeal and the stakes. Tax authorities see efficiency gains; privacy advocates see risks to sensitive data. Germany's proposal forces a direct choice between the two.
For government professionals involved in tax administration or data governance, AI for Government resources can help assess how AI systems affect workflows and compliance. Understanding AI Data Analysis practices is also relevant to evaluating how real versus synthetic training data affects system accuracy and fairness.
Your membership also unlocks:
