Hi Friends,

Even as I launch this today ( my 80th Birthday ), I realize that there is yet so much to say and do. There is just no time to look back, no time to wonder,"Will anyone read these pages?"

With regards,
Hemen Parekh
27 June 2013

Now as I approach my 90th birthday ( 27 June 2023 ) , I invite you to visit my Digital Avatar ( www.hemenparekh.ai ) – and continue chatting with me , even when I am no more here physically

Translate

Friday, 30 October 2026

One Certificate, Many Lenses – AI Regulation Across Sectors

 To: 

Shri Ashwini Vaishnaw, Hon. Minister for Electronics & IT

Cc: 

Shri Anil Kumar Lahoti, Chairman, TRAI; 

Secretariat, AI Governance & Economic Group (AIGEG)

cp@trai.gov.in /  secretary@meity.gov.in  /  moeit@gov.in  /  moeit@gov.in

=================================================

Subject: 

One Certificate, Many Lenses – avoiding conflicting AI rules across sectoral regulators

===================================================

Dear Shri Ashwini Vaishnawji,


TRAI has just urged telecom operators to test AI models before deployment, log their decisions and keep humans able to intervene. These are sound principles. But they apply equally to RBI, SEBI, IRDAI, CDSCO, DGCA, CERC and every other sectoral regulator.

If each regulator now writes its own AI rulebook, the same AI model will face a dozen different audits, vocabularies and logging formats. Worse, the rules will sometimes pull in opposite directions.

In my blog below, I propose a five-layer architecture called One Certificate, Many Lenses. It builds on my earlier proposals, Parekh's Law of Chatbots (Feb 2023) and UNARAI (May 2023), and on TRAI's own 2023 call for a common AI framework across all sectors.

I request that this be considered as an input to MeitY's forthcoming AI consultation, and that AIGEG take up the idea of an AI Regulators' Council.

With regards,

Hemen Parekh

www.HemenParekh.ai / www.IndiaAGI.ai / 31 Oct 2026

Blog: One Certificate, Many Lenses

The news

TRAI is pushing for responsible use of AI in telecom networks, to make telcos accountable. TRAI Chairman Anil Kumar Lahoti has asked operators to invest in responsible AI, carry out risk assessments, keep logs of decisions and test models before deployment (Daily Pioneer). He also wants humans to keep the ability to monitor, audit and intervene as networks become more autonomous (Asia News Network).

The question it raises

Telecom is one industry with one regulator. India has many such industries, each with its own regulator: RBI, SEBI, IRDAI, CDSCO, DGCA, CERC and more. All of them are already using AI.

So shall each regulator issue its own, separate AI guidelines? And if they do, won't they end up working at cross purposes?

There is a further twist. Telecom networks carry the Internet, through which AI travels. But so do WiFi, enterprise LANs and satellites, none of which TRAI fully covers.

Where regulators will collide

TRAI's principles (test before deployment, log decisions, keep a human override) are not telecom-specific at all. That is exactly why a dozen separate rulebooks will clash.

  1. One AI system, two regulators. A telco's AI spam filter that wrongly blocks bank OTPs is a TRAI problem and an RBI problem at once. Each may demand a different audit format for the same model.
  2. One model, many sectors. A single foundation model running customer service for a bank, a hospital and a telco would need certification three times, under three vocabularies.
  3. Conflicting data rules. One regulator's data-retention mandate can collide with DPDP's data-minimisation principle.
  4. The pipe is the wrong place to regulate. AI reaches citizens over fibre, WiFi and satellite alike. Regulating it through the carrier creates gaps and duplication at the same time.

The gap is already visible. A July 2026 commentary noted that India's new apex AI body, AIGEG, does not include the sectoral regulators who have done the most detailed AI work (USISPF).

I had flagged this risk earlier this year in Keeping AI Options Open (May 2026): sectoral rules bring precision, but raise complexity and compliance cost for multi-use AI systems.

My solution: One Certificate, Many Lenses

Layer 1 – Certify the AI once, nationally

In Parekh's Law of Chatbots (Feb 2023), I proposed that every chatbot be submitted to an approval authority (IACA) and released only after receiving a certificate: P for public use, R for restricted use. In my 1st Amendment (June 2026), I added N (No Release) and D (Destroy).

Applied to this problem: a single national AI Certification Authority certifies each model once. Pre-deployment testing, decision-log formats and incident reporting become common standards. The certificate travels with the model into every sector.

TRAI itself came close to this in 2023, recommending an independent AI authority, a risk-based framework and a common regulatory framework covering all sectors (Digital Watch).

Layer 2 – Sector regulators write only "use-case annexures"

  • TRAI, RBI, SEBI and others add requirements specific to their domain, such as network-autonomy limits for telecom or algorithmic-trading limits for markets.
  • They may make the base certificate stricter, never weaker.
  • They may not redefine its terms.

This keeps their domain expertise without a dozen parallel rulebooks.

Layer 3 – An AI Regulators' Council, on the GST Council pattern

This follows the federal-council design of my Central Testing Commission proposal.

  • Members: all sectoral regulators plus MeitY, sitting under AIGEG.
  • Lead-regulator rule: for a cross-sector system (a telco filter blocking bank OTPs), one regulator leads and the others concur.
  • Single window: one compliance filing and one shared AI incident registry, reused by every regulator.
  • Conflict register: contradictory rules are logged and resolved within a fixed timeline.

Layer 4 – Regulate the driver and the vehicle, not the road

TRAI should govern AI that runs the network itself: routing, fraud detection, spam filtering. It should not govern the AI services travelling over that network. Those fall under the Layer 1 certificate and the relevant sector annexure, whether they arrive over fibre, WiFi or satellite.

Layer 5 – Plug into the world through UNARAI

AI crosses borders as easily as it crosses networks. In United Nations Agency for Regulating AI (UNARAI) (May 2023), I proposed a UN agency of all member states to develop and enforce AI standards. India's national certificate should be built for mutual recognition abroad, the way ISO and CE marks work today.

The other side

Some scholars argue that AI poses very different challenges in different domains, so one regulator cannot fit all of them (Ryan Nabil, NTUF). That concern is fair, and it is exactly why my design keeps sector annexures instead of handing everything to one super-regulator.

Closing

India does not need one AI regulator, nor a dozen competing ones. It needs one certificate, read through many lenses, with a council to settle disputes. Then a model certified once can serve a bank, a hospital and a telco without being re-judged three times.

With regards,

Hemen Parekh

www.HemenParekh.ai / www.IndiaAGI.ai / www.HemenParekh.in / 31 Oct 2026

Posted by

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)