After CCTV reported on malicious code hidden in open-source agent projects, a tax official from Chongqing wrote about how tax departments should adopt AI. He advocates three principles: choose scenarios carefully, using AI for audit case selection, risk scanning and document proofreading, but not letting AI directly issue conclusions that affect taxpayer rights; manage permissions strictly, requiring human confirmation for operational actions and granting backend data access on a least-privilege basis; and control external output, sticking to official wording for policy explanations and routing uncertain questions to humans rather than letting AI fabricate answers. The core idea is to keep AI working within a compliant and controllable boundary while safeguarding data security.
News Detail
Amid the Open-Source Agent Craze, Tax Authorities Need a Clear AI Strategy
Following security alerts about the OpenClaw open-source agent project, an official from the Chongqing tax authority's information center wrote that tax departments need a clear AI strategy, not a simple use-or-not choice. The article proposes three principles: choose application scenarios carefully, using AI for audit case selection, risk scanning and document proofreading but not for directly generating conclusions affecting taxpayer rights; manage authorization strictly under least-privilege rules with human confirmation for operational actions; and control output boundaries, sticking to official language for policy explanations and routing uncertain questions to human channels to prevent AI hallucinations from causing misinterpretation.
2026-08-27