India's telecom regulator has proposed rules for artificial intelligence in telecom, requiring operators to adopt board-approved AI policies, rate every AI system by risk and report critical and high-severity AI incidents to the Telecom Regulatory Authority of India (TRAI). The draft, released Oct. 30, is open for comment until Nov. 27.

The rules would reach operators serving a vast market: India had 1.37 billion telephone subscribers at the end of September, TRAI data showed. For investors in Indian telecom stocks, including the newly listed Jio Platforms, the draft adds a compliance layer at a time when operators are building AI into network management and customer service. It also means AI vendors selling to Indian telcos would face new due-diligence checks and contract terms.

What TRAI's AI guidelines would require of telcos

The guidelines would apply to any company holding a licence or authorization to provide telecom services or run telecom networks that uses, or plans to use, AI, according to the draft document.

Each such operator would have to classify its AI systems as high, medium or low risk, based on how much they matter to the network, to financial and operational results and to consumers. No AI system could be used unless it sits in an up-to-date inventory and has been approved by a designated officer. The internal AI policy would have to be filed with TRAI within seven working days of taking effect and reviewed at least once a financial year.

Operators would also need an AI Governance and Safety Committee, approved by the board and headed by a senior executive designated as the AI Governance and Safety Officer. AI incidents would be graded in four tiers, from low to critical severity, logged in a central database and investigated. The two most serious tiers would go to TRAI "in a prescribed form, manner, and timeframe," the draft said, though it does not yet set that timeframe.

Before deployment, systems would face conformity checks covering accuracy, explainability, fairness, drift resilience and human oversight. The draft points operators to Telecommunication Engineering Center standards for fairness testing and for recording AI incidents.

The draft cites the India AI Governance Guidelines the central government released in February, saying TRAI wants consistent application of those principles in telecom. It adds no new penalties. Violations involving AI would be handled under existing laws, rules and licence conditions, the document said.

Telecom operators carry the risk for vendor AI

The provision with the widest commercial reach is accountability. Operators would answer for the outcome of every AI system they use, whether built in house, bought from a vendor or a mix of both, the draft said.

That means due diligence on suppliers before purchase, covering the vendor's credibility, the system's limits and the quality of its training data. Supplier contracts would need enforceable terms on audit and inspection rights, incident management and risk allocation, and equipment makers would have to disclose any AI embedded in what they sell.

MediaNama reported that the rule would also cover the AI-based spam filters TRAI ordered operators to deploy for calls and text messages in 2023, so operators would be accountable when those filters err. The outlet set the draft against Anthropic's Oct. 26 U.S. listing filing, which says a substantial portion of its traffic falls under zero-data-retention terms and that gaps in audit trails could impair its ability to investigate incidents or respond to regulators.

BusinessLine reported that the rules could raise compliance spending for operators over the medium term while opening business for AI governance, cybersecurity, testing and audit firms. An unnamed industry veteran told the newspaper that recent cybersecurity incidents had made the government want telcos to understand the AI systems that sit behind phone banking and enterprise services.

India's wider AI rulebook takes shape

TRAI's move came the same week as a second federal step on AI. The Ministry of Electronics and Information Technology (MeitY) said on Oct. 31 it had released a handbook on its Data Empowerment and Protection Architecture for AI training, a framework meant to let institutions pool data for model training while keeping control of it, according to The Economic Times' ETCIO. IT Secretary S. Krishnan unveiled the handbook, developed with the iSPIRT Foundation, at a roundtable in Bengaluru, the report said.

India still has no single AI law. Most AI-related oversight runs through the IT Act of 2000 and IT rules, Business Standard reported on Oct. 8, when IT Minister Ashwini Vaishnaw said the government would publish a draft consultation paper on AI regulation within 30 days. He said India needed a "techno-legal approach" and that industry would carry much of the regulatory load, the newspaper reported.

MediaNama reported on Nov. 2 that the ministry was still working on those wider rules. Vaishnaw's 30-day window runs to about Nov. 7, and TRAI's comment period closes Nov. 27.