TRAI lays out guidelines for responsible AI adoption in telecoms - Developing Telecoms

India's Telecom Regulatory Authority has published draft guidelines for "responsible AI adoption" in telecom networks, infrastructure and services — three nouns that between them cover every cable, mast and app in the country, so full marks for scoping. The substance, as Developing Telecoms reports it, is a risk assessment for "any AI implementation" and a "set time limit" for reporting critical incidents. Note which words are doing the heavy lifting: "any", "set", and the conspicuous absence of an actual number. The deadline exists in principle, which is the most comfortable place for a deadline to exist.

This is the same genre our old friends in the US Senate were practising with the PATIENT AI Act: the document that offers ideas about a technology rather than rules for it. TRAI has one advantage — it actually regulates networks, rather than merely gesturing at reimbursement rates for health AI. But "draft" is the operative word in both capitals, and a draft guideline is a press release with a consultation period attached. "Responsible adoption" is what everybody says immediately before adopting irresponsibly and then pointing out that the risk assessment was filed correctly.

The risk assessment is the loveliest artefact in the document: every operator assesses the risk of its own AI, presumably against criteria it has a hand in writing, and files the result somewhere tidy. And the critical-incident clock — the one element with genuine teeth — is a time limit whose length has been deferred, so the first properly critical incident will almost certainly land before anybody agrees how late the report is allowed to be. It is a start, and it is being reported as one, which is more than most regulators manage. Give it a couple of consultation rounds and the deadline will acquire a number, the form will acquire a portal, and the AI will carry on doing precisely what it was going to do anyway.