TRAI: Proposes responsible AI guidelines, seeks comments

The Telecom Regulatory Authority of India (TRAI) has proposed responsible AI guidelines for telecom networks and services, seeking written comments by 27 November 2026. Its draft, dated 30 October 2026, would require covered operators to adopt board-approved artificial intelligence policies, assess systems before deployment and report critical- and high-severity AI incidents to the regulator.

The proposals would place AI oversight within operators’ corporate governance, procurement and consumer-complaint processes. TRAI’s draft calls for human supervision of AI systems and a senior executive to head an AI Governance and Safety Committee. These are consultation proposals, not requirements presented as already in force.

TRAI says AI could improve telecom efficiency but also introduce risks through inadequate checks before deployment or insufficient supervision of systems that learn and interact autonomously. The draft identifies unintended bias, changes in model behaviour and adversarial attacks as risks to telecom infrastructure, particularly its critical components.

Which operators and systems would be covered

Under TRAI’s draft, the guidelines would cover entities licensed under the Telegraph Act, 1885, or authorised under the Telecommunications Act, 2023, to provide telecom services. They would also cover entities authorised under the latter Act to establish, operate, maintain or expand telecom networks.

In both categories, applicability would depend on whether the entity uses, or intends to use, AI in its networks or service provision. The proposed scope therefore includes planned deployments, rather than only systems already operating.

Each covered entity would need an internal AI policy approved by its board of directors or equivalent governing body, according to the draft. The policy would have to be reviewed at least once each financial year and whenever an additional review becomes necessary. Operators would file the policy, and subsequent amendments, with TRAI within seven working days of their taking effect.

TRAI also proposes a board-approved AI Governance and Safety Committee, headed by a senior executive designated as the AI Governance and Safety Officer. Its membership would draw from relevant functions such as network operations, service delivery, information technology, data management, regulatory affairs and consumer affairs.

Risk checks before deployment

TRAI proposes classifying every AI system as high-, medium- or low-risk. The assessment would consider the system’s importance to networks or service delivery, its financial and operational effects, and its potential consequences for consumers.

That classification would help determine oversight. Under the draft, a designated officer would approve a system before use, with approval powers delegable to suitable officers according to risk. Operators would review each system’s risk category at least once a financial year or in response to specific AI incidents.

The draft would also require a comprehensive inventory covering active, inactive and decommissioned AI systems. Each entry would identify the system’s source, purpose and risk category. No system could be used unless it appeared in that inventory.

Before deploying a new system or modifying an existing one, operators would undertake validation, conformity assessment and impact assessment, TRAI proposes. In practical terms, these are checks on how the system performs, whether it meets relevant requirements and what effects its use could have.

TRAI’s proposed checks cover accuracy, robustness, explainability, fairness, resilience to model drift and human oversight. Model drift refers to changes in a model’s behaviour or performance over time; explainability concerns whether its operation or outputs can be understood. For fairness assessment, the draft identifies the Telecommunication Engineering Center’s standard on fairness assessment and rating of AI systems, or an applicable international standard, as options where appropriate.

Supplier contracts would address AI risks

TRAI’s draft extends the proposed checks to AI obtained from third parties. Before acquisition or use, operators would examine the supplier’s credibility, the system’s soundness and limitations, and the suitability and quality of its training data.

Suppliers and prospective suppliers would also have to disclose AI embedded in telecom equipment, systems or solutions being procured, under the proposed framework. This would bring embedded AI within the operator’s assessment process rather than limiting scrutiny to separately purchased AI products.

The draft calls for enforceable supplier-contract provisions covering regulatory compliance, audit and inspection rights, incident management, risk allocation and accountability before deployment.

Within the operator, TRAI proposes periodic AI-related training for technical, operational, regulatory and business personnel involved in human oversight. The draft also requires lawful and secure handling of data throughout the lifecycle of deployed systems, including collection, use, sharing and disposal, in accordance with applicable privacy, confidentiality, cybersecurity and other statutory requirements.

Incident reporting and consumer complaints

TRAI proposes a separate classification for AI incidents: critical, high, moderate and low severity. Operators would record all incidents in a central database, including those arising during testing, and follow predefined procedures to mitigate disruption.

Under the draft, every incident would be investigated and analysed, with the analysis shared across relevant units of the organisation. Critical- and high-severity incidents would be reported to TRAI in a prescribed format, manner and timeframe.

The reporting provision does not specify a numerical deadline. ThePrint, in its report on the draft, also notes that TRAI has not set out the reporting format or deadline, or defined what would qualify as a critical- or high-severity incident.

For consumers, TRAI proposes a predefined grievance mechanism to resolve reported AI-related harm within a “reasonable timeframe”. Operators would maintain a central database recording those reports and their resolution. The consumer-protection clause does not assign a fixed number of days for resolving them.

How the draft fits the wider AI framework

TRAI says the proposals seek to implement the Government of India’s India AI Governance Guidelines consistently within telecom. The national framework promotes innovation alongside safeguards, recommends voluntary measures to encourage compliance and allows additional risk-mitigation obligations in specific contexts, including sensitive applications. It also envisages sector regulators continuing to exercise enforcement powers.

For its own mandate, TRAI’s draft cites Section 11(1)(b)(v) of the TRAI Act, 1997. As described in the document, that provision concerns setting and ensuring telecom quality-of-service standards to protect consumers.

The regulator’s examination of AI predates this draft. Its consultation paper dated 5 August 2022, titled “Leveraging Artificial Intelligence and Big Data in Telecommunication Sector”, discussed AI use in network operations, customer services and infrastructure operations, alongside risks such as privacy violations and biased results. The October 2026 draft sets out proposed operator-level controls for those technologies.

TRAI says stakeholders’ written comments on the draft will be posted on its website. It invites submissions, preferably electronically, to its Advisor for Networks, Spectrum and Licensing by 27 November 2026.