New Delhi: Telecom operators will have to report critical and high-severity failures of artificial intelligence (AI) systems in networks to the Telecom Regulatory Authority of India (TRAI), and will be held accountable for the outcomes of every AI system they use, according to draft guidelines released Friday by the regulator.
The draft states that an operator “shall be accountable for the outcome of all AI systems used by it”, be it in-house built, procured from a third party, or a combination of both.
Currently, AI is used to detect spam and fraud. Bharti Airtel launched a network-based AI spam detection system in September 2024. Built in-house by its data scientists, the system labels calls and SMSes as ‘Suspected SPAM’ using a proprietary algorithm.
As of January 2026, Airtel said the system had flagged more than 71 billion calls and 2.9 billion SMSes as spam. A system of this kind, which decides how a call is shown to a subscriber, would fall within the scope of the draft.
Incident reports to TRAI must cover material malfunctions, recurring failures and the steps taken to mitigate them. The regulator said it may use this information for sector-wide learning, risk assessment, advisories and “other lawful regulatory purposes”.
The TRAI does not define what counts as a critical-severity, or high-severity incident. It also does not set out the format or deadline for reporting. It says these will be “prescribed”, without saying when or how. Each operator will define its own incident categories in an internal policy.
Feedbacks are to be submitted by 27 November.
Also Read: Anthropic study’s new AI cautionary tale: AI agents can influence expert opinion to serve bias
Mandatory language for a voluntary framework
The Ministry of Electronics and Information Technology (MeitY) has said it is not regulating AI, and its guidelines are voluntary, with seven sutras adapted from the Reserve Bank of India’s FREE-AI report.
The ministry is relying on existing laws and sector regulators instead of a new AI statute. The framework calls for encouraging compliance through voluntary measures, with additional obligations possible in specific contexts such as sensitive applications.
The TRAI’s draft uses “shall” throughout. For instance, operators “shall” adopt an internal AI policy approved by their board, and “shall” file it with TRAI within seven working days of it coming into effect. They “shall” set up an AI Governance and Safety Committee headed by a senior executive, designated the AI Governance and Safety Officer. They “shall” keep records to show compliance and hand them over to TRAI on demand.
Neither does the draft contain a penalty clause. It simply says any act or omission involving an AI system that breaks an existing law, licence condition or authorisation will be dealt with under that law.
For its legal basis, the authority cites Section 11(1)(b)(v) of the TRAI Act, 1997. This provision empowers it to lay down quality-of-service standards to protect consumers.
Who it covers
The norms will apply to holders of a licence under the Indian Telegraph Act, 1885, or an authorisation under the Telecommunications Act, 2023, for providing telecom services. They will also cover entities authorised to set up and run telecom networks. Both categories are covered only if they use or intend to use AI.
Operators will have to classify every AI system as high-, medium- or low-risk. The classification is based on the system’s importance to the network, its effect on financial and operational outcomes, and its impact on consumers. No AI system can be used unless it is on a central inventory and has been cleared by a designated officer.
Before deploying a new system, or modifying an existing one, operators must carry out validation, conformity and impact assessments. These cover accuracy, robustness, explainability, fairness and resilience to model drift.
Vendor contracts
Operators will have to run due diligence on third-party AI before buying it given that they rely on foreign equipment vendors for most of their network gear and software.
The watch-out list includes checking the credibility of the supplier and the quality of the data used to train the system. They must also obtain disclosures from vendors about AI embedded in telecom equipment. Supplier agreements will need enforceable clauses on audit and inspection rights, incident management, risk allocation and accountability.
Users will get a grievance route for AI-related harm. Operators must keep a central database of such complaints and resolve them within a “reasonable timeframe”, according to the draft. It, however, does not specify that timeframe.
TRAI’s earlier position
The telecom regulator had earlier sought a broader regulator for AI. In July 2023, it recommended an independent statutory authority, the Artificial Intelligence and Data Authority of India (AIDAI) for ensuring development of responsible AI and
regulation of use cases in India.
It also proposed a risk-based framework under which high-risk use cases directly affecting people would face legally binding obligations.
The technology industry opposed creating new laws or a statutory body. In August 2025, TRAI chairman Anil Kumar Lahoti said the recommendations were under government review.
The government did not create a new authority. Instead, its framework left enforcement with sector regulators, supported by an AI Governance Group and an AI Safety Institute. Friday’s draft is TRAI’s first set of AI rules within that structure, limited to the operators it already regulates. Operators must review their AI policies and the risk categories of their systems at least once a financial year.
(Edited by Tony Rai)