Telangana Issues Strict AI Guidelines For Tax Officers To Protect Taxpayer Confidentiality

The Commercial Taxes Department of the Government of Telangana has issued Circular No. 1/2026, setting strict regulatory boundaries for state officers utilizing artificial intelligence and third-party web platforms during official duties. The directive, dated August 18, 2026, balances operational efficiency with data protection, establishing firm protocols regarding taxpayer confidentiality and statutory compliance.

While the department encourages responsible AI integration to boost the speed and consistency of official workflows, the new mandates explicitly forbid compromising data security or bypassing independent professional judgment. Officials routinely manage highly sensitive materials, including corporate returns, bank particulars, audit records, e-way bills, and investigation files, making data exposure a critical administrative vulnerability.

The department highlighted that confidentiality requirements under Sections 152, 158, and 158A of the Telangana Goods and Services Tax Act, 2017, remain legally binding across all field offices. Taxpayer information gathered under statutory authority cannot be shared with private commercial AI service providers under the guise of administrative convenience. Such disclosure violates state frameworks and the Digital Personal Data Protection Act, 2023.

Unauthorised transmission of taxpayer details to external digital tools risks severe administrative fallouts. The circular warns that exposing confidential records to public models can trigger departmental disciplinary proceedings under Telangana Civil Services rules, alongside potential criminal liabilities and vulnerabilities to external legal challenges, damage claims, and adverse judicial commentary.

A central feature of the new operational framework is the strict assignment of personal liability directly to the handling officer. Officials cannot deflect responsibility by claiming that outsourced personnel, data-entry operators, stenographers, or personal assistants uploaded the records. Every breach is treated as the primary officer's direct administrative failure.

the department underscored the necessity of independent application of mind in quasi-judicial duties. Notices, adjudication orders, discrepancy reports, and appellate decisions must reflect an officer's own reasoned interpretation of evidence and law, warning that unverified reproduction of AI text undermines statutory validity and leaves administrative orders legally vulnerable.

Officers face cautions regarding the tendency of artificial intelligence models to generate fluent yet factually incorrect material, including non-existent case laws, incorrect statutory citations, or fabricated factual details. Consequently, every legal citation and notification proposed for use in official orders must undergo independent verification from authoritative primary sources before signing.

The regulatory framework imposes an absolute prohibition on typing, scanning, photographing, or transmitting any taxpayer particulars—such as GSTIN, PAN, bank accounts, or investigation records—into unauthorized platforms. Additionally, employees are banned from connecting departmental backend databases, such as GSTN, the e-way bill system, or BIFA, to external application programming interfaces or browser plugins.

Notwithstanding the strict prohibitions, the guidelines outline permissible scenarios where artificial intelligence may support administrative tasks. Officers may leverage AI for generic legal research, abstract inquiries into points of law, and refining the grammar, structure, or clarity of draft documents, provided no identifiable taxpayer data or case specifics are included.

To ensure complete institutional compliance, the department mandated that the circular be served upon every staff member, requiring signed and dated acknowledgements of receipt. Supervisory officers must also ensure that historical taxpayer data previously entered into third-party tools is deleted, chat histories are cleared, and model-training retention settings are disabled across all official and personal devices.