Cybersecurity leader Trend Micro has announced a new joint initiative with NVIDIA aimed at safeguarding the next generation of Agentic AI systems. The collaboration focuses on integrating security deep within the hardware and software layers of what NVIDIA calls “AI factories”—the vast data centers that power modern AI models.
The announcement, made in India on October 29, 2025, outlines a comprehensive approach to protecting AI infrastructure, from the physical hardware all the way up to the application layer. The goal is simple yet ambitious: help organizations deploy AI systems faster and with much lower risk.
Key Takeaways
- Hardware-Level Security: Trend Micro’s Trend Vision One platform will now run directly on NVIDIA BlueField DPUs (Data Processing Units).
- Application-Level Safety: The system integrates with NVIDIA NeMo Guardrails to regulate AI model behavior and prevent misuse or abuse.
- End-to-End Protection: Designed specifically for large-scale AI factories, the system supports compliance with strict industry and government standards.
What is Agentic AI?
Agentic AI refers to AI systems capable of more than just responding to inputs. These systems can perceive their environment, make independent decisions, and take actions toward achieving goals. While that autonomy opens exciting new possibilities, it also brings novel security risks that traditional defenses aren’t built to handle.
Securing the AI Foundation
At the heart of this new security offering lies the NVIDIA BlueField DPU—a specialized processor that offloads networking, storage, and security tasks from the main CPU.
By embedding Trend Vision One Endpoint Security directly onto the BlueField DPU, Trend Micro can observe and secure data center operations at the hardware level. This “agentless” setup allows for real-time threat detection and response without compromising AI application performance.
Interestingly, this capability forms part of a reference design for the NVIDIA AI Factory for Government, ensuring that it meets the rigorous compliance and security standards of federal agencies and highly regulated sectors.
“As enterprises deploy AI factories, they need to secure large-scale, high-speed infrastructures without slowing innovation,” said Ofir Arkin, Sr. Distinguished Architect for Cybersecurity at NVIDIA. He added that running Trend Vision One on BlueField “establishes a new class of endpoint detection and response for AI factories,” combining hardware-level isolation with advanced threat intelligence.
Protecting the AI Application Layer
Hardware is only one side of the equation. As AI systems become more autonomous, they face threats that target their behavior rather than just their code or data. Attacks like prompt injection (where malicious inputs trick AI into harmful actions) or agent abuse (where AI agents are manipulated to perform unintended tasks) are becoming increasingly common.
To address these risks, Trend Vision One’s AI Application Security (AI Guard) now integrates directly with NVIDIA NeMo Guardrails—a framework that enforces safety rules for Large Language Models (LLMs).
This combination gives security teams powerful control to:
- Define and enforce safety policies for AI models
- Detect and block issues like data leakage, hallucinations, jailbreaks, and prompt injections
- Continuously monitor AI interactions for abnormal behavior
- Automate threat response actions
“Agentic AI promises to create business agility, but only if we build it on secure foundations,” noted Sharda Tickoo, Country Manager for India & SAARC at Trend Micro.
The integrated system operates on three main pillars: content moderation, hardware isolation, and data protection. It filters harmful outputs, leverages BlueField’s isolation features for resilience against attacks, and includes encryption and zero-trust mechanisms to maintain compliance with privacy regulations like GDPR and CCPA.
Related FAQs
What is Agentic AI?
Agentic AI systems (or AI agents) are more autonomous than traditional chatbots. They can perceive their digital environment, make independent decisions, and execute complex tasks through tools or APIs—essentially acting as digital assistants capable of real-world action.
What is an NVIDIA BlueField DPU?
A DPU (Data Processing Unit) is a specialized processor designed for data centers. It manages networking, storage, and security workloads, freeing up CPUs and GPUs to handle the compute-heavy AI model operations.
How does this solution prevent prompt injection attacks?
It uses a dual-layer defense strategy. The NVIDIA BlueField DPU provides hardware isolation to contain threats, while NeMo Guardrails detects and blocks malicious input patterns before they reach the AI model.
What is an AI Factory?
An AI factory is the full-scale ecosystem—hardware, software, and data—used by organizations to build, train, and run AI models at scale.
Who is this solution for?
The Trend Micro–NVIDIA collaboration is targeted at large enterprises, cloud providers, and government agencies operating AI factories, especially those bound by strict compliance requirements or high-security needs.
Final Thoughts
This partnership between Trend Micro and NVIDIA marks a meaningful step toward making Agentic AI not just more capable, but genuinely safer to deploy. It’s a recognition that as AI grows more powerful, security can no longer be an afterthought—it has to be part of the architecture itself. And perhaps, that’s the only way to keep innovation and safety moving in step.
