Google’s AI Push in India Meets a New Era of Agentic Models

When Google stepped onto the stage at I/O Connect India 2026, the message was clear: artificial intelligence is no longer a research demo, it’s becoming a daily utility for teachers, farmers, doctors and every‑day shoppers across the country. The company didn’t just talk about slick chat‑bots; it rolled out concrete tools aimed at keeping India’s digital infrastructure safe as AI agents start to act on their own.

“The agentic era extends that responsibility,” a Google India spokesperson said in a press release. “Software can now interpret intent, use tools and take action autonomously. Safety can’t be an after‑thought; it has to be baked into the architecture from day one.” That philosophy is behind three flagship initiatives that Google is now piloting with trusted government bodies and large enterprises such as Flipkart.

First up is Sec‑Gemini v3, a specialised cybersecurity agent that can sift through massive streams of security logs, spot anomalies and help incident‑response teams understand threats in minutes instead of hours. Built on the lessons learned from Google DeepMind’s “Big Sleep” vulnerability‑research agent, Sec‑Gemini can even flag emerging exploits before they hit the wild.

Second, Google highlighted CodeMender, an autonomous coder that generates patches for discovered vulnerabilities and pushes them directly to open‑source repositories. The tool made headlines in July 2025 when it was credited with thwarting an active exploit in the wild – a first for an AI‑driven security agent.

Third, the company opened the door to a more defensive development culture by releasing CAPSEM (Capabilities Security for Agents) as open‑source. Think of it as a sandboxed virtual machine where AI agents live, only able to touch the resources you explicitly allow. If an agent is compromised, the rest of the system stays insulated.

Beyond the tech, Google is pushing for industry‑wide standards. It is championing the Device‑Bound Session Credentials (DBSC) proposal at the W3C, a cryptographic way to tie login tokens to a user’s physical device, rendering stolen cookies useless. Alongside that, the Agents‑to‑Payments (AP2) protocol aims to make sub‑$100 transactions executed by AI agents both secure and auditable.

While Google is busy fortifying India’s AI ecosystem, another major development rippled through the community. Mira Murati, the former OpenAI CTO who briefly steered the company during its 2023 leadership shake‑up, launched Thinking Machines Labs and unveiled its first model, Inkling. Described as a Mixture‑of‑Experts transformer with a staggering 975 billion parameters (41 billion active at any moment), Inkling can chew through a context window of up to one million tokens and was pre‑trained on 45 trillion multimodal tokens.

Inkling isn’t just a behemoth for show; a lighter sibling called Inkling‑Small (12 billion active parameters) is also being released, promising strong performance at a fraction of the cost and latency. By publishing the model’s weights openly, Murati signals a shift toward more transparent, fine‑tunable AI—something that could dovetail nicely with Google’s push for secure, accountable agents.

The convergence of Google’s security‑first rollout in India and the arrival of open, massive models like Inkling suggests we’re standing at a crossroads. On one side, powerful agents capable of independent action demand robust safeguards; on the other, the democratisation of model access promises unprecedented innovation for developers everywhere. For India, a nation where AI‑driven solutions can boost agriculture yields, improve rural healthcare and personalise education, the stakes have never been higher.

What remains to be seen is how quickly regulators, startups and multinationals can align on standards, share best practices and, crucially, keep the technology safe for the billions of users it will soon touch. One thing is certain: the ink is still drying on this new chapter, and the words being written could well define the next decade of AI in the subcontinent.