At Reliance’s recent annual general meeting, Akash Ambani, the Managing Director of Jio Platforms, announced an AI agent that can be triggered with the “Hey Jio” keyword while you are on a call with someone.  

“No app to download, no number to add, available to every Jio customer in every Indian language. Just say ‘Hey, Jio.’ And your AI agent joins the call for as long as you like, always, and only with your consent.” —Akash Ambani, MD of Jio Platforms.

Watch Akash Ambani’s announcement, including the recorded video demo of the AI agent, from 38:00 to 41:00 through this time-stamped URL.

He said the AI agent can do the following things: 

• Transcribe, summarise, and list action items discussed. 
• Identify multi-speakers in conference calls (up to eight); 
• Do tasks like ordering food, booking a cab, reserving a table, or scheduling a meeting discussed during the call.

Here are a few questions we have for Jio: 

1. Since Jio’s network-level AI agents are triggered by the ‘Hey Jio’ keyword, it might be required for it to process each spoken word. So, does it mean that Jio will continuously listen and process each spoken word on every call on its network?

If the keyword detection runs at the device level (locally), then no audio would have to leave the device unless or until the triggering keyword is detected. Such privacy-respecting architecture should be ideally used. However, Jio explicitly said that their AI agent is built “into the heart of the Jio network.”

How does this work at the network level without necessarily processing each word? If this is the behind-the-scenes mechanism, i.e., every conversation is subject to AI processing, then it’s clearly a privacy violation if the purpose limitation principles are not followed under the DPDP Act.

This gives another reason to rely on end-to-end-encrypted internet calls over cellular calls. 

Read more about how Amazon’s Alexa handles your data here: [ Original | Archived ] 

This brings us to the question: Is my call data being used to train AI models right now?

Reliance Jio’s Privacy Policy doesn’t specifically mention anything about AI data training. However, it reads that non-personal information will be used to analyse trends and learn user behaviour. Here is the archived version of its privacy policy. Read Privacy Policy here: [ Original | Archived ]

2. If I am the unconsented user on the network, during the call, if the other person invokes the AI agent, Akash Ambani says that the other user will be notified. This is okay. But, will I have an option to disable the AI agent on the call?

Because consent can’t be one-way when it involves two or more individuals. It has to be on both sides. And, moreover, notification shouldn’t be deemed as consent. It is the announcement/declaration, not a question of consent. Under India’s DPDP Act, consent should be free, specific, informed, and unambiguous. Otherwise, one-sided consent would be deemed to be self-protection of the company from liability of privacy and consent violation and not a pure users’ privacy-respecting architecture. 

I’d also argue that the other person on the call should have the right and ability to waive off AI agents if they wish. The user may have the right to record/document at the device level. In this case, the data isn’t locally stored; the data travels to AI servers, possibly abroad. Would a government official discussing national security be comfortable if they knew that the call would be processed by foreign AI companies, which may have partnered with their government or defense department? 

Quick note on the US govt and AI companies: The US government’s Department of Defence already has at least eight companies on its classified network. These are: Claude, Google, Oracle, SpaceX, OpenAI,, NVIDIA, Reflection, Microsoft, Amazon Web Services,

Therefore, it is very important for the public to know the data processing terms when AI agents are deployed at the network level.

3. Which AI/LLM model processes the calls? Jio’s own (if it has one), Google Gemini, Meta Llama, or any third-party provider?  

Right now, Jio has a partnership with Google, bundling its Gemini AI Pro through certain recharge packs. On the other hand, Meta has a 9.9% stake in Jio Platforms since 2020 and also partnered to build a 168 MW AI data center in Gujarat’s Jamnagar. 

So, who will be processing the call data? If transcription and summarisation tasks are handled through Google’s Gemini API, then call data would be transmitted to Google’s servers and potentially subject to Google’s own data processing terms, not necessarily Jio’s own privacy policy. The same argument applies to Meta’s AI models as well.

Also, what would be the data retention period of both data stored and data generated? Will that data be repurposed for AI data training? 

Quick note on Reliance’s lack of interest in LLMs: In the October 2025 earnings call, Reliance Industries Limited (RIL) clarified that developing LLMs is not their focus; rather, they wanted to concentrate on AI infrastructure, such as data centers. After a few months, they announced the Meta-Reliance AI data center in Jamnagar. 

4. Since this feature is built into the network and not just an app, does the user have a meaningful way to permanently opt out? And, more importantly, what happens when someone who never consented but gets called by a Jio user who has?

 If a non-Jio user, let’s say, someone from Airtel, BSNL, or a foreign telco, receives a call from a Jio user who invokes the AI agent, then it might mean that person has agreed to Jio’s terms of service. But does it also mean that the non-Jio user is also considered to have taken the consent since he is on a call with a Jio user? How will Jio design this consent mechanism? 

If the non-Jio has no direct relationship with Jio’s policies, yet their voice is being identified, transcribed, summarized, and retained by either Jio’s or Jio’s affiliated AI systems, how would the DPDP Act’s data principal and data fiduciary principles apply here? 

5. Jio’s AI agent apparently “knows your usual outlet and saved address,” which can be used to execute tasks delegated during the call. Is/will Jio collect behavioural data too across the calls to build this user profile? If so, under what terms?

Memory is the core feature for an AI agent’s efficient functionality. So, while using user data for building user profiles is one potential concern, there might be competition issues as well. For instance, when a user says to order a pizza to their home address, as shown in the demo video, it might repurpose my address-related data for this task. That’s fine. But which platform will it choose? Which outlet will it choose? How would the human-in-the-loop be designed at the network level for acceptance and denial of AI agentic decisions? However, most of these concerns under this question are also valid for other AI agents, too, not just Jio’s network-level AI agent.

At this moment, Jio’s announcement of a network-level AI agent raises more questions than the answers it solves. We gave a briefer version of these questions to Jio. We’ll update the copy if they respond to us.

Recommended read: TRAI’s Recommendations on Leveraging Artificial Intelligence and Big Data in Telecommunication Sector — [ PDF | Archived PDF | MediaNama coverage ]

Also Read:

• Jio Platforms:9 AI Developments Disclosed Ahead of IPO 
• Reliance Launches AI Arm, Partners With Google and Meta 
• JioHotstar Expands AI Team, Builds New AI Division