Spain's AI Tax Secrecy Plan Tests Transparency and Taxpayer Rights

Spain is proposing an amendment to its Artificial Intelligence Bill that would modify Article 116 of the General Tax Law to restrict access to information about the algorithms and AI tools used by the Spanish Tax Agency for selecting taxpayers for inspection. This move has implications for AI governance, transparency, and the balance between data-driven efficiency and taxpayer rights.

What happened

The proposed amendment would limit taxpayers' ability to learn how the Spanish Tax Agency uses AI to select individuals for audit. According to the OECD, more than 70% of tax administrations now use AI in compliance management and taxpayer services, with machine learning and generative AI central to risk assessment and audit selection. Spain's proposal has drawn strong opposition from the Association of Tax Advisers (AEDAF), which argues it reduces transparency and accountability. The international law firm Amsterdam & Partners contends the amendment may conflict with the EU AI Act and the General Data Protection Regulation (GDPR), as well as broader constitutional guarantees of transparency and due process.

The debate extends beyond the amendment itself. Spain's digital tax infrastructure has faced scrutiny over the use of Huawei storage systems for taxpayer data, raising questions about data security and the safeguards surrounding information exchanged under international tax treaties. For comparison, the UK's HMRC CONNECT system reportedly held more than 55 billion taxpayer-related data points by 2023, combining public and private data for compliance work.

Why AI builders should care

For teams building AI systems for government or regulated sectors, this case highlights the growing demand for explainability and auditability. Opaque algorithms erode public trust and invite legal challenges under frameworks like the EU AI Act. Builders need to design for transparency from the start, ensuring model decisions can be documented, explained, and contested. The Spanish proposal is a concrete example of how governments may attempt to shield AI systems from scrutiny, which could set a precedent for other jurisdictions.

Practical implications

Developers working on tax compliance AI or similar high-stakes applications should expect stricter requirements for model documentation, audit trails, and human oversight. The EU AI Act already mandates transparency for high-risk AI systems, and Spain's move could accelerate enforcement. Teams should invest in explainability tools, maintain clear records of training data and decision logic, and build interfaces that allow human reviewers to override or challenge automated decisions. The Huawei storage concerns also underscore the importance of data residency and security in government AI deployments.

Caveats

The amendment is proposed, not yet law. The exact legislative text is not publicly available in the source material. The analysis from Amsterdam & Partners is a legal opinion, not a court ruling or official government position. The OECD data reflects a broad survey of tax administrations, not specific to Spain. The Huawei storage concerns are separate from the AI amendment but part of the broader data infrastructure debate. Readers should monitor the legislative process and consult official sources for precise legal language.