Technical Context
I went straight to the source because headlines about an "AI tax" are off the mark. This isn't about a new levy—it's an amendment to Spain's AI bill that touches Article 116 of the General Tax Law and how the tax authority uses algorithms to select individuals and companies for audits.
For me, that's where real AI integration starts, not PR. When a government builds AI-based risk scoring, it's effectively making decisions with legal consequences. The immediate question isn't "Is the model smart?" but "Can you later explain why you were flagged?"
According to Euractiv and legal commentators, the controversial part is the plan to severely restrict access to information about these systems. Critics are blunt: it would create an almost opaque shield around the algorithms, blocking them not only from taxpayers but sometimes even from external oversight.
I find the conflict at the rule-architecture level compelling. On one hand, the tax office needs anti-fraud measures and protection of its models from abuse. On the other, the EU AI Act and the logic of GDPR generally push the system toward transparency, explainability, and the ability to challenge an outcome when automated logic affects the rights of a person or business.
And yes, the Spanish tax agency previously stated publicly that automated actions wouldn't be based solely on AI output. But if the selection mechanism for audits is hidden, the right to an explanation becomes highly conditional. Formally, a human is in the loop, but in reality you still can't see how you ended up in the risk funnel.
Impact on Business and Automation
For companies, this is a bad signal not just as taxpayers but also for those implementing AI in regulated processes. If the government sets an example of opaque automation, the private sector will start pushing for black boxes where they shouldn't.
The winners here are those who prioritize efficiency over auditability. The losers are businesses that have to defend themselves against decisions whose logic is hidden, and teams that later have to untangle the legal risks of such AI automation.
I'd view Spain as a test case for the whole of Europe. If this model passes, disputes about where anti-fraud protection ends and procedural rights violations begin will quickly spread to other countries.
If your AI solution development is hitting compliance, explainability, or contentious automated decisions, let's break it down at the process and access-rights level. At Nahornyi AI Lab, I work hands-on with these bottlenecks: you can build AI automation that accelerates work instead of creating a new legal minefield.