As India’s AI buildout gathers pace, Operant AI has launched a new partnership program aimed at a part of the stack that is becoming impossible to ignore: real-time security at the inference layer. The San Francisco-based company announced its AI Infrastructure Ecosystem Partnership Program today in Bengaluru, positioning it as a way to embed security directly into the infrastructure running enterprise AI workloads in production.
The timing is deliberate. India’s AI infrastructure story is moving quickly, with GPU deployments, data-centre expansion, and new inference platforms reshaping how enterprises plan to deploy AI systems at scale. But alongside that momentum, security concerns are rising just as fast.
Operant AI’s pitch is straightforward: if AI systems are becoming more autonomous, more connected, and more deeply embedded in sensitive environments, then protection cannot remain an afterthought. It has to sit where the model actually runs.
The new partnership program is open to AI infrastructure providers, inference platforms, and MCP-compatible application vendors. The focus is on integrating Operant’s real-time defence tools directly into customer environments, especially as more enterprises move from AI experimentation to production-scale deployment.
Operant AI: India’s AI expansion is creating a new security challenge
The announcement comes against the backdrop of a rapid expansion in India’s AI infrastructure.
Operant AI said that under the IndiaAI Mission, the government has already deployed 38,000 GPUs, with another 20,000 GPUs recently announced at the India AI Impact Summit 2026. Across public and private deployments combined, the company said more than 80,000 GPUs are now in operation in India.
The company also pointed to broader infrastructure growth across the country. Data-centre operators such as ST Telemedia Global Data Centres, CtrlS, Sify Technologies, Nxtra by Airtel, and Princeton Digital Group are expanding capacity for AI workloads, while GPU cloud providers including Yotta Data Services, NxtGen Datacenter & Cloud Technologies, and Neysa Networks are building specialised AI compute platforms.
Operant AI added that India’s total data-centre capacity is estimated to rise from roughly 2 GW today to 8–9 GW by 2030, driven by major investments from hyperscalers and global infrastructure players including Google, AWS, and Reliance.
That growth, however, is also widening the gap between infrastructure scale and security preparedness. The company cited the FICCI–EY Risk Survey 2026, released in February, which found that 61% of Indian business leaders now see cyber-attacks and data breaches as posing significant financial and reputational risks to their organisations.
In other words, India’s AI infrastructure race is no longer only about capacity. It is increasingly about how securely that capacity can be put to work.
Why inference-layer security is getting attention
Operant AI argues that traditional security approaches are not keeping up with the way enterprise AI systems now operate.
That matters because many AI deployments are no longer limited to isolated models answering prompts. They are evolving into agentic systems that can access tools, handle sensitive information, and connect with external systems. As these systems become more autonomous and increasingly linked through standards such as the Model Context Protocol (MCP), the potential attack surface expands as well.
The company says its platform is built to discover, detect, and defend AI models, agents, and MCP-connected systems while they are actively running inference. Instead of scanning only at the perimeter, the idea is to monitor the actual runtime flow of requests and responses, and stop threats before they affect model behaviour or downstream systems.
Among the risks it highlights are prompt injection attacks, tool-level vulnerabilities, and runtime manipulation, threats that are increasingly relevant as production AI deployments become more complex.
New GPU-accelerated security tools
As part of the launch, Operant AI introduced GPU-accelerated versions of AI Gatekeeper and MCP Gateway. The company said these are designed to provide high-speed security scanning and validation for models, agents, tools, and the wider inference-layer toolchain without creating latency bottlenecks.
The company outlined four core capabilities of its inference-layer security approach:
- Prompt injection detection and blocking, aimed at stopping malicious prompts from altering model behaviour, extracting training data, or bypassing safety controls
- Zero-day attack protection, using behavioural monitoring to detect unusual querying patterns or suspicious agent-tool interactions
- MCP security for agents, with runtime validation of agent actions and tool calls
- Intelligent rate limiting, intended to reduce unnecessary model invocations and protect against costly token consumption caused by attacks or inefficient agent behaviour
The emphasis here is not just on security as a compliance checkbox, but on security as an operational requirement for enterprise AI at scale.
Operant AI ties the launch to India’s enterprise AI moment
Commenting on the launch, Vrajesh Bhavsar, Co-founder and CEO of Operant AI, said the rapid scale and autonomy of AI systems have moved beyond the capabilities of older security controls.
He said the company believes the inference layer is where security must now be enforced, especially as Indian enterprises deploy AI models and agents across sectors such as financial services, healthcare, and the public sector. Bhavsar added that the company’s GPU-accelerated AI Gatekeeper and MCP Gateway, combined with the new ecosystem program, are designed to meet the speed and protection requirements of production AI systems in India.
Operant AI also used the announcement to spotlight Bhavsar’s background. According to the company, he previously built core security technologies for iOS and macOS at Apple, including Data Protection and the Secure Enclave, and later founded the Machine Learning business unit at ARM, where he helped build an ML ecosystem program with more than 50 partners.
Tenstorrent partnership adds infrastructure angle
The company also highlighted its collaboration with Tenstorrent, linking the security announcement to the broader push toward faster inference infrastructure.
Aniket Saha, VP of Product Strategy at Tenstorrent, said that as AI moves toward always-on agents, infrastructure must remain both high-performing and open by design. He said the partnership with Operant AI would allow Tenstorrent customers to pair high-throughput inference platforms with real-time visibility at the inference runtime layer.
Operant AI said the integration would place its monitoring and enforcement capabilities directly in the inference stack on top of Tenstorrent hardware, giving joint customers visibility, compliance support, and traceability while using high-performance inference infrastructure.
A bet on security as a differentiator
For infrastructure providers, this launch is also about market positioning.
As enterprises become more cautious about the security posture of the AI platforms they adopt, providers may increasingly need to show that their environments are not only powerful, but secure enough for real-world deployment. Operant AI is effectively betting that runtime defence at the inference layer will become part of that competitive equation.
The company said partners in the new program will gain access to products including AI Gatekeeper, MCP Gateway, and Agent Protector, allowing them to offer integrated real-time security to customers. It also said it is the only vendor featured across six Gartner AI security reports, including the AI TRiSM Market Guide, API Protection Market Guide, MCP Gateways Innovation Insight, and How to Secure Custom-Built AI Agents.
At a moment when India is scaling AI infrastructure at speed, Operant AI’s announcement reflects a broader shift in the market: enterprises are no longer asking only how fast AI can run. They are also asking how safely it can run once it is live.
Also Read: Pronto: Instant Home Cleaning Services in Just 10 Minutes!
