A 2023 paper in Tax Research takes on a problem that is easy to skip: smarter AI for tax administration needs more and better data, but the more real and concentrated that data is, the wider the door for leakage and misuse. The author proposes synthetic data on three conditions - strict neutrality, strict de-identification and a minimum data scale. Models can still be trained while real records stay put, and liability becomes traceable. The paper warns that high-risk public-sector AI in tax is most exposed to data poisoning and algorithmic discrimination, and argues from an overall national security standpoint that synthetic data both lowers economic risk and keeps data and network risks in check.
News Detail
Applying Synthetic Data in the Digital Transformation of Tax Administration
A 2023 paper in Tax Research argues that smarter AI for tax administration demands more and better data, while existing studies focus on data authenticity and silos and underplay security. Synthetic data, built on strict neutrality, strict de-identification and a minimum data scale, makes training safer and separates AI from real records, clarifying legal responsibility. The author stresses that high-risk public-sector AI must guard against data poisoning and algorithmic discrimination, and that from an overall national security perspective synthetic data lowers economic risk while keeping data and network risks contained.
2026-11-03