“What we definitely need is a very light-touch progressive enabling regulatory system. We need an accountability system which does not seek transparency into systems but seeks transparency and demands the highest quality of service and excellence to what we deliver to our subscribers,” said Ambika Khurana, Chief Regulatory & Corporate Affairs Officer, Vodafone Idea, at the Cellular Operators Association of India’s (COAI) event on AI and Telecom.
Unlike MeitY’s short-lived 2024 plans for mandatory pre-deployment AI testing, the Department of Telecommunication (DoT) and telecom operators are now pushing for a risk-based regulatory approach. This should focus on service quality and built-in human oversight rather than invasive reviews/audits of proprietary AI systems. For some additional context, in June 2026, DoT announced a National Working Group (NWG) on AI-related technical standardizations. Read more about it here.
“As earlier Ambika said, the government approach is a light-touch approach. And it’s not that we regulate much, and then we hinder the development and the tech developments, specifically. So, this is the approach the government has taken, and the Ministry of Electronics and Information Technology has proceeded with the India AI mission, about 10,000 crores over a period of 5 years. And there is an India AI Security Institute also under that,” said Mehul Joshi, Deputy Director General & CISO at DoT.
Absolute humans in the loop may not be possible, but a framework for accountability needs to be created: “There was a lot of discussion on humans in the loop and humans taking the final decisions. I agree, not possible to get totally humans in the loop, but an obvious oversight to ensure that the right design, the right architecture framework, and oversight are maintained so that there is absolute alignment with the right principles, there is right accountability, and that’s where a lot of regulatory clarity comes into play,” she further said.
“We’ve been saying that you need to have something called an ‘agentic Security Operations Center (SOC).’ When agents are attacking you, a single human being cannot be able to take up that problem. You will have agents defending you as well,” ideated Pankaj Verma, National Cyber Security Advisor at Cisco.
We need observability into agentic actions: “Now, these agents, what are they really doing? How many tokens are they using? Tokenomics become very, very important. Which agent is the one which is hogging the most tokens? Can I do something? Everything has to be done on the LLM, or can I start seeing something at the LLM level? That’s the agent observability that really needs to come in in a big way,” said Verma.
“We’ve been discussing we can’t have human in the loop. I would also agree, but definitely we need human on the loop. You can’t just say that the agent did it. Who’s going to take accountability for that agent? So guardrails, definitely,” he reasoned.
Nations should analyze the AI systems before deployment in real world: “Every nation should strive to have a sovereign understanding and capability to analyze and interrogate these systems before we deploy them in our economies and in our security apparatus,” said Joe Joscelyne, a UK official at the British High Commission in India. He referred to the UK’s AI Security Institute that analyzes the frontier models.
What DoT is doing: “Telecom Technology Development Fund (TTDF), which is being funded by the Department of Telecom. In this, more than 20 R&D projects are being undertaken. The scheme overall has an outlay of Rs. 500 crore. Many projects are going on when security, trust, and digital resilience and 5G advanced and 6G networks,” he added. He specified the following examples:
- IIT Indoor, that is doing a project called as a Darpan, AI-assisted, reliable, Indigenous Digital Twin for sixth Sense 6G Networks and digital resilience.
- “IIST Trivindram, which is AI for supporting integrated sensing and secured communication. A 6G-assisted edge network focuses on AI for securing 6G edge networks.”
- “MNIT Jaipur is doing another one, which is decentralized deep learning for 6G communication.
- “Apart from the initiative of the Department of Telecom, we also have the India AI Mission that is supporting more than 13 projects in bias mitigation, machine unlearning, privacy-preserving AI, algorithmic auditing, and explainability,” he claimed.
Speakers skipped MeitY’s question: Citing the above-mentioned MeitY’s now stalled idea of auditing AI systems for security before real-world deployment, a participant at the discussion asked, “What do you think that these AI models or generative AI models need permission or approval from the MeitY before they are deployed in India to have this security, and deserve national security?” Speakers skipped answering this question and quickly moved to a “telecom-related” question.
Also Read:
