TRAI's New A2P Call Rules, WhatsApp Charges and Claude Haiku 5.5: AI Pulse

TRAI has finalised new A2P call rules that change how businesses can place automated calls in India. On 18 September 2026 TRAI issued the Telecom Commercial Communications Customer Preference (Third Amendment) Regulations, 2026, which require every business using application-to-person (A2P) calls to declare it to its telecom provider, and let operators use AI to flag spam senders.

If you run outbound calling, robocalls or an AI calling agent, this is the change to plan for this month. Below are the TRAI rules, the new WhatsApp charges that started on 1 October, and three AI tool updates from the past fortnight.

1. TRAI's A2P call rules: what changed

The press release from TRAI and the matching PIB release on 18 September list these changes:

• Declare A2P calls. TRAI says "every entity using A2P calls must pre-declare such use to its TSP" (your telecom service provider). TRAI says undeclared A2P calls "will be treated as UCC", meaning unsolicited commercial communication.
• AI flagging of senders. Operators will use AI to spot numbers likely to be sending spam. If five or more numbers linked to one sender are flagged within ten days, the sender faces KYC re-verification or disconnection.
• Fewer complaints needed. Action can follow 3 unique complaints in ten days instead of 5, where AI flagging backs them up.
• Termination charge. The receiving operator may charge the calling operator up to ₹0.05 a minute on these calls.
• Old consent must be provable. Legacy consent counts only if it was "obtained through verifiable means."
• 140 and 1600 series protected. Caller ID and spam apps may not blanket-block or spam-tag calls from the 140xx, 1600xx and 1601xx series.

When the rules apply

The regulation text says most provisions start 30 days after publication in the Gazette. The A2P declaration rule and the termination charge start after 60 days, and the complaint changes after 90 days.

Onmanorama and the Daily Pioneer reported in late September that the rules had been published in the Gazette. We could not confirm the exact Gazette date, so treat late October as the earliest the 30-day rules apply, and late November for the 60-day rules.

What it means for AI calling agents: a voice agent that places outbound calls from software is likely to count as an A2P call. Ask your telecom or calling provider how they will handle the declaration, which number series your calls should use, and how they record consent. If you are budgeting for an AI calling setup, our AI calling agent cost guide covers the other line items.

2. WhatsApp service messages now cost money

Meta's non-template message pricing page says: "Effective October 1, 2026, Meta will charge on a per-message basis for service messages." Utility template messages sent in reply to a customer are now charged too.

Gupshup, a Meta partner, says Meta gives a free monthly tier of 1,000 service messages per business phone number. Business Standard reported on 1 October that additional service messages in India cost ₹0.115 each, and that the marketing rate stays at ₹0.8631.

Using those two figures, here is the monthly bill for service replies from one number:

• 1,000 replies: ₹0 (inside the free tier)
• 5,000 replies: 4,000 × ₹0.115 = ₹460
• 20,000 replies: 19,000 × ₹0.115 = ₹2,185

The amounts are small. The bigger risk is delivery: Make's 2 October update notes that once the 1,000 free messages are used, further messages will not be delivered unless a valid payment method is set in Meta's Billing Hub. Check that before your support bot goes quiet in the middle of a month. Our WhatsApp automation page covers how we set up support flows.

3. AI tool updates worth knowing

Claude Haiku 5.5 cuts the cost of simple AI tasks

Anthropic released Claude Haiku 5.5 on 7 October at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Anthropic's release notes also say Max and Team plans now include monthly API credits for running your own apps and agents.

Google Cloud's Gemini agent

Google Cloud's Gemini at Work post on 8 October introduced the Gemini agent, "a universal agent for work." Google says it "chooses the best model for the job" and "has built-in cost controls." The post does not give pricing.

Make adds AI fallback and prompt caching

Make's 9 October release lets you set a fallback AI provider for its AI Agent app, which Make retries once if the main provider is down. It also adds prompt caching so agents reuse instructions and tool definitions, and Make says both are on all plans.

What changed and what to do this month

Three of this week's updates point the same way: send cheap work to a cheap model and keep a backup ready. Saturday's deep dive shows how to design an AI agent that picks the right model for each task, with real per-task costs worked out. For outbound calling that stays within the rules, see how our AI voice agent works.

Get posts like this in your inbox once a week. Subscribe to NXT Weekly.

Keep reading