F5 & NVIDIA tie up on AI guardrails for production

F5 has integrated F5 AI Guardrails with NVIDIA NeMo Guardrails to centralise policy enforcement across AI applications.

The integration is aimed at organisations running multiple AI models and frameworks across distributed technology environments, where security controls and governance often sit inside individual applications rather than in a shared layer.

F5 software will inspect prompts and large language model responses, while NVIDIA NeMo Guardrails remains the framework used to apply safety and security rules within AI applications. The approach lets organisations use NVIDIA's framework while handling inspection and policy enforcement separately.

That structure addresses a common issue for companies moving AI systems from pilot projects into production. Security teams often want tighter oversight of prompts, outputs and data exposure, while developers want to avoid rewriting applications each time a new control is introduced.

Under the setup, security policies can be managed centrally instead of being embedded in each application. This allows organisations to inspect AI traffic, apply security requirements and audit activity without changing application code.

Production focus

The product is intended to protect production AI systems by screening prompts and responses between users and applications. It is designed to help reduce risks including prompt injection, exposure of personally identifiable information, data leakage and harmful outputs.

The integration also offers a single operational view across AI applications, models and frameworks. That is intended to give security teams one place to monitor behaviour and apply policies across different business units.

Kunal Anand, Chief Product Officer at F5, said many companies do not lack AI tools but consistency in how they secure them.

"Enterprises do not have a shortage of AI applications. They have a shortage of consistent security and governance across them," said Kunal Anand, Chief Product Officer at F5.

"As AI moves into production, fragmented controls create risk, complexity, and delay. This integration gives security teams a unified point to inspect AI traffic, enforce policy, and govern AI applications across their environments, while giving developers the freedom to keep building."

NVIDIA described the integration as an extension of its existing guardrails framework for customers that want additional layers of runtime inspection around large language model activity.

"NVIDIA NeMo Guardrails provides an open, programmable framework for applying safety and security policies to AI applications," said Ash Bhalgat, Senior Director of AI Networking and Security Solutions, Ecosystem and Marketing at NVIDIA.

"The integration with F5 AI Guardrails expands the range of protections customers can use to secure LLM prompts and responses as they move AI agents into production."

Broader ecosystem

Several partner companies also backed the announcement, reflecting broader market interest in tools that govern AI workloads as they spread across private infrastructure, cloud systems and mixed computing environments.

Equinix linked the offering to the need for consistent policy enforcement across distributed infrastructure used for AI. Red Hat said the integration adds runtime protections for customers deploying AI workloads through its joint work with NVIDIA. WWT said the technology strengthens governance and security checks before and during deployment of AI systems.

The integration works regardless of the underlying large language model, meaning customers can change models or expand into new business units without altering the security layer. That separation may appeal to large organisations seeking common governance standards while allowing internal teams to choose different AI tools.

F5 presented the architecture as one in which the framework library, microservices, orchestration tools and security controls can evolve independently. In practice, that means development teams can keep building AI applications while security teams maintain separate control over inspection, policy and governance.

For companies adopting AI agents and other generative AI systems, that division of responsibilities has become as much an operational issue as a technical one. Many businesses have rushed to test new models, but production rollouts have exposed problems with oversight, auditability and the ability to apply the same standards across many applications.

By placing inspection between users and applications, F5 is positioning its software as a shared enforcement layer rather than a feature tied to one model or application stack. NVIDIA NeMo Guardrails, meanwhile, remains the programmable framework organisations can use to standardise AI safety and security rules.

The integrated offering is now generally available.