The Telecom Regulatory Authority of India published draft guidelines on responsible artificial intelligence in networks that would force carriers to secure board approval for internal AI policies, classify each deployment by risk, and maintain living inventories of models that touch customer service or core switching.
What carriers must document
Under the consultation, released as India's larger AI governance conversation accelerates at the Ministry of Electronics and Information Technology, telecom operators and other licensed network entities would categorize systems as high-, medium- or low-risk depending on operational impact and consumer harm potential. High-risk tools would need documented human oversight, pre-deployment safety assessments and senior executive committees empowered to shutter malfunctioning models.
TRAI also proposes that operators cannot outsource liability simply by buying models from vendors. If a chatbot misroutes emergency calls or a fraud-scoring engine blocks legitimate subscribers, the licensee remains accountable to the regulator regardless of where the weights were trained.
Overlap with MeitY and CERT-In
The draft comment window runs until 27 November, parallel to but separate from MeitY's forthcoming economy-wide consultation paper and CERT-In's security-auditing guidance for AI developers. Telecom is receiving early sectoral treatment because networks already run automated traffic management, robocall filtering and customer-support assistants at scale.
Industry lawyers note that TRAI's inventory requirement resembles European risk registers but stops short of banning specific use cases outright. Instead, the emphasis is on governance processes: annual policy reviews submitted within seven working days of board sign-off, central databases of incidents, and consumer complaint logs tied to automated decisions.
Consumer-facing stakes
Indian mobile users increasingly interact with operator apps for recharge, complaint ticketing and fraud alerts. Those touchpoints are ripe for large-language-model assistants, yet telcos have been cautious after early hallucination incidents in test environments. The draft's approval gate — no AI in production without a designated officer's sign-off — aims to slow reckless rollouts without freezing experimentation in labs.
What changes for enterprises
Enterprises leasing private 5G or managed WAN services from the same carriers may see contractual pass-throughs requiring disclosure when AI optimises bandwidth. Cloud majors selling edge compute into telco data centres could face questionnaire expansions during security audits.
Comments due in late November will test whether Bharat Sanchar Nigan Ltd, Reliance Jio and Bharti Airtel push back on committee staffing costs or embrace the rules as a competitive signal of trust. Either way, India's telecom layer is now an explicit front in the country's AI rulemaking, not a spectator waiting for a general statute.
Sector consultations
Bharti Airtel and Vodafone Idea have piloted network-energy optimisation models that automatically sleep radio layers during low traffic. Those tools would likely count as medium-risk under TRAI's taxonomy because they affect service availability even if consumers rarely notice the change.
Reliance Jio's consumer chatbots already triage recharge complaints; expanding them to billing disputes would trigger the draft's human-oversight clauses. Product managers must document fallback paths to live agents when confidence scores fall below thresholds yet to be defined in the final rules.
Comparison with global telcos
European carriers already publish AI transparency summaries for roaming fraud detection. Indian draft rules align in spirit but emphasise board accountability more than public algorithmic impact statements. Multinationals operating here may harmonise policies across regions if TRAI finalises inventory formats compatible with GSMA templates.