TRAI spam sharing plan faces AI model mismatch, consent gaps

Telecom operators cannot agree on who a spammer is because their AI detection systems differ, a MediaNama roundtable on August 12 heard. The Telecom Regulatory Authority of India (TRAI) has directed operators to share AI-detected spam signals across networks within two hours from February 2026, but each operator uses different AI models and parameters. A number flagged as spam by one network may not be flagged by another, raising concerns about acting on 'a guess about a guess.' Operators argue shared signals should inform, not dictate, blocking decisions, and resist passing AI indicators to law enforcement as a flag is only a probability.

The roundtable also highlighted that the consent system promised under TRAI's 2018 telecom rules was never implemented. The Distributed Ledger Technology platform lacks a consent layer, so messages cannot be checked against what a person actually agreed to receive. Speakers said consent today is often broad or manufactured, and earlier attempts like the Digital Consent Acquisition mechanism failed as businesses feared customer fatigue. A recent pilot of a revocable consent system has worked, but is not yet operationalised. Meanwhile, spam is shifting to WhatsApp and RCS, raising questions whether TRAI's rules should apply to internet platforms.