The Federal Tax Service Talks About AI—But It's More About Automation Than Intelligence

The Federal Tax Service regularly states that it already uses artificial intelligence: to review tax returns, predict risks, identify business fragmentation, and uncover complex schemes. This is confirmed by official representatives of the service—from Elita Elkin, head of Interdistrict Tax Inspectorate No. 29 for the Sverdlovsk Region, to Tatyana Rybalko, head of the Federal Tax Service for the Smolensk Region.

It sounds impressive. But if we break down exactly what lies behind the wording, the picture becomes less rosy.

FNS publications state that the algorithm “analyzes documents,” “identifies potential errors,” and “predicts tax risks.” This is a classic description of a rule-based system: rules, threshold values, and statistical models. Not a neural network, not an LLM, not an agent-based architecture—but automation that existed long before the hype surrounding AI.

The problem isn’t that automation is bad. On the contrary: for tax compliance, it is effective and necessary. The problem lies in the terminology. When any form of automation is called “artificial intelligence,” it blurs the concept to the point of complete meaninglessness. And it creates the false impression that tax authorities are already working with GPT- or Claude-level models that can reason on their own about tax evasion schemes.

There’s an important lesson here for developers and engineers: the market (and the government as part of it) is rebranding existing technologies as AI on a massive scale. If you’re building a product with real LLM agents, your users will compare you not to actual counterparts, but to the marketing hype where “AI” means any rule with an “if” statement.

The next question is: when will the Federal Tax Service actually start using language models to analyze tax schemes—and what will that mean for businesses? That hasn’t happened yet.

Source: All Articles in a Row / Artificial Intelligence / Habr