# India Just Told AI Voice Agents to Declare Themselves. Undeclared Calls Are Now Spam.

URL: https://zippiai.com/blog/india-just-told-ai-voice-agents-to-declare-themselves-undeclared-calls-are-now-spam
Author: ZippiAi Team
Published: 2026-09-24T02:25:00.423Z

**On Friday India's telecom regulator became the first major market to regulate AI voice agents by how a call is started rather than what the voice says. Three regulators now use three different tests. Only one of them touches inbound support.**

Two days after Instinct and Meta's Muse started phoning US businesses on their users' behalf, the Telecom Regulatory Authority of India (TRAI) amended its commercial-communications rules to deal with the mirror image of that problem: businesses phoning people with software. The amendment, reported by TechCrunch on 18 September, does two things. It forces caller-ID apps such as Truecaller to push their users' spam reports into a blockchain-based platform run by the telcos. And, in a section that got far less attention, it redraws the line between a human call and a machine call.

Under the new rules, "calls made automatically, without a person directly dialing the number, will now fall under TRAI's application-to-person (A2P) framework." That covers robocalls, prerecorded messages and calls made with artificial voices, which in 2026 means AI voice agents. Companies using such systems must "declare their use and the phone numbers involved to their telecom operators in advance." Undeclared A2P calls, TRAI said, will be treated as spam. Telcos may also levy a termination charge of up to 5 paise (about 0.052 US cents) per minute on A2P traffic, with certain designated number ranges exempt.

No commencement date was given, and TRAI did not answer TechCrunch's questions about enforcement. But the shape of the rule is clear enough to plan around, and the scale of the market it applies to is not in doubt.

## **Why India, and why now**

India is the largest voice market on earth that also has a spam problem worth regulating. Truecaller, the Stockholm-based caller-ID company, reported in February that its Indian users encountered around 42 billion spam calls in 2025, of which it blocked nearly 12 billion. India accounts for well over 350 million of Truecaller's 500 million-plus monthly active users. Any rule that changes how automated calls terminate on Indian networks is, by volume, the biggest voice-AI rule anywhere.

It is also the first rule to define the machine call by initiation rather than by voice. Sumeysh Srivastava of The Quantum Hub, who leads the firm's telecom-regulation work, told TechCrunch the key test "is how a call is initiated, rather than simply whether it uses an AI-generated voice." That distinction matters more than it sounds, because the two other regulators that have acted on AI voice chose different tests, and the three rules land on contact centres in three different places.

## **Three regulators, three tests**

  

India: TRAI amendment (18 Sep 2026)

United States: FCC declaratory ruling (8 Feb 2024)

European Union: AI Act Article 50 (applies from 2 Aug 2026)

What triggers the rule

The call was not directly dialled by a person (robocall, prerecorded, artificial voice)

The call uses an "artificial or prerecorded voice", which the FCC confirmed includes AI-generated voices

An AI system interacts directly with a natural person in a genuine two-way exchange

Who it binds

Companies making the calls, and the telcos terminating them

Callers initiating outbound calls to consumers

Providers and deployers of the AI system

What you must do

Declare the use and the phone numbers to your telco in advance

Obtain prior express consent before calling (written consent for telemarketing); existing identification and opt-out rules apply

Inform the person, clearly and at the start of the first interaction, that they are talking to AI, unless it is obvious to a reasonably well-informed person

What happens if you don't

The call is treated as spam; telcos may charge up to 5 paise/min on A2P traffic

The call is illegal under the TCPA; state attorneys general can pursue it

Fines up to €15 million or 3% of worldwide turnover

Inbound support calls affected?

No. The customer dialled the number, so the call is person-initiated

No. The ruling attaches to calls the business initiates

Yes. A voice agent answering an inbound call is a direct interaction and must identify itself

Grey zone

Software-assisted calls where a human is still involved, such as contact-centre dialling and click-to-call

Whether AI that merely selects which prerecorded message to play is caught (the FCC says yes)

The "obvious" exception, which the Commission says must be read restrictively

Read across the bottom two rows and the practical picture emerges. If you run a voice AI that answers the phone, India's new rule does not apply to you, the FCC's ruling does not apply to you, and the EU's rule applies to you in one specific way: say so at the start. If you run a voice AI that makes calls, all three apply, in different ways, and India's is the one that adds a per-minute cost and a registration step.

## **The grey zone is where contact centres actually live**

The clean inbound-outbound split has a complication that Kazim Rizvi, founding director of the New Delhi think tank The Dialogue, put to TechCrunch directly. The new definition "could also cover calls made using software even when a person is still involved, such as calls from contact centers and click-to-call services." His warning: "Without that distinction, the A2P category risks becoming broader than the regulatory harm it is intended to address."

Most enterprise outbound traffic is exactly this. A predictive or progressive dialler places the call; a human agent speaks. A customer taps "call me back" on a website; software originates the call and connects a person. On a strict reading of "without a person directly dialing the number", both are A2P and both need to be declared. On a purposive reading they are not what the rule is aimed at. Until TRAI says which reading it intends, the conservative assumption for anyone operating in India is that any call your software originates should be on the declared list, whether or not a human ends up on it.

There is a second-order effect for inbound operations as well. The consumer agents that started calling businesses this week are, by TRAI's definition, A2P callers: no person dialled the number. Instinct Concierge and Muse are US-only today. When an equivalent reaches India, the burden of declaring falls on the agent's operator, not on the business receiving the call. But the business still has to answer it, and a DTMF menu has no way to ask a calling agent whether it is registered. A conversational front door at least gets a transcript.

## **What this changes for a voice AI programme**

For outbound: the cost model moves. A termination charge of 5 paise a minute is small per call and material at scale, and it lands on the A2P category specifically. Voice-agent campaigns in India will need to be justified against a per-minute network cost that a human agent's dialler call may not carry, depending on how the grey zone resolves. The registration requirement also means the numbers a voice agent calls from become a compliance asset to be inventoried and kept current, in the same way sender IDs already are for SMS.

For inbound: the EU has set the only obligation that bites, and it is a disclosure obligation, not a consent or registration one. The Commission's guidance is that the notice must come "from the start of the first interaction in a clear and distinguishable manner." A voice agent that opens with "you're speaking to an automated assistant" satisfies it; one that is designed to pass as a person does not, and the "obvious" exception is explicitly to be read narrowly.

For analytics: every one of these rules is, underneath, a record-keeping rule. TRAI wants to know which numbers are automated. The FCC's regime turns on whether consent existed. The EU's turns on whether the disclosure was made. An operation that logs each automated call with its origin (inbound or software-initiated), the identity disclosure, the consent basis where relevant and the outcome has the evidence for all three. An operation whose only record of an automated call is a billing CDR does not.

## **What to check this quarter**

Three questions for any organisation running or buying voice AI. First, list every call your systems originate without a human dialling, including click-to-call and dialler traffic, and decide now whether each category would be declared under a TRAI-style rule; the exercise is useful even outside India, because the initiation test is the one most likely to be copied. Second, listen to the first five seconds of your inbound voice agent and ask whether a reasonably well-informed caller would know it is AI; if the answer depends on the caller's sophistication, the EU test is not met. Third, confirm that your call records can answer, per call, "was this machine-initiated, was AI disclosed, and what happened," because that is the audit trail all three regulators will eventually ask for.

_At ZippiAi we build Aura, a voice AI companion and analytics layer for enterprise support. Aura answers the phone; it does not run robocall campaigns, so the rules above land on it mainly as a disclosure and record-keeping question, which is how we think they should be designed: every conversation logged with who started it, what the caller was told, and how it ended._

  

* * *

**Sources**

  

*   TechCrunch, 18 Sep 2026 (published 19 Sep 01:00 UTC): [India forces caller-ID apps to feed spam reports to telcos](https://techcrunch.com/2026/09/18/india-forces-caller-id-apps-to-feed-spam-reports-to-telcos/)
*   TechCrunch, 17 Sep 2026: [Rival AI agents, Instinct and Meta's Muse, both add the ability to make calls](https://techcrunch.com/2026/09/17/rival-ai-agents-instinct-and-metas-muse-both-add-the-ability-to-make-calls/)
*   FCC news release, 8 Feb 2024: [FCC Makes AI-Generated Voices in Robocalls Illegal](https://docs.fcc.gov/public/attachments/DOC-400393A1.pdf) (Declaratory Ruling FCC 24-17, CG Docket No. 23-362)
*   Mayer Brown, 27 Feb 2024: [FCC Declares Authority and Intent to Regulate AI-Generated Calls under the TCPA](https://www.mayerbrown.com/en/insights/publications/2024/02/fcc-declares-authority-and-intent-to-regulate-ai-generated-calls-under-the-tcpa)
*   European Commission, updated 24 Jul 2026: [Transparency obligations under Article 50 of the AI Act (FAQ)](https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act)