Nvidia launches new NIM microservices in NeMo Guardrails
The AI hardware and software vendor introduces new constraints to make sure AI agents and models don't go off topic, leak personal data or become vulnerable to jailbreaks.
Nvidia added new NIM microservices to NeMo Guardrails to help enterprises address trust, safety, security and compliance problems with AI agents.
Nvidia NeMo Guardrails is a platform for defining, orchestrating and enforcing AI safety protections. First introduced in 2023, the platform enables users to apply multiple constraints on AI models depending on business needs.
Constraints include preventing the model from veering into unsafe content, keeping it on topic, preventing jailbreaks and preventing the sharing of personal data.
New NIM microservices
Now, NeMo Guardrails will include new NIM microservices, released on Thursday, to guide AI agents. The new microservices include the content safety NIM, topic control NIM and jailbreak detection NIM.
The content safety NIM prevents AI models and agents from generating harmful outputs. It was trained using the Aegis AI Content Safety Dataset, a human-annotated data source owned by Nvidia but publicly available on Hugging Face.
The topic control NIM microservice keeps conversations on approved topics. The jailbreak detection NIM microservice keeps outsiders from jailbreaking the AI agent or model.
"The key is that we need to keep AI agents on track while also making sure that they're fast and responsive to interact with other AI agents and also end users," said Kari Briski, vice president of enterprise AI models, software and services at Nvidia.
The introduction of the new microservices comes a week after Nvidia introduced its world foundation model aimed at moving AI from the digital world to the physical world. Nvidia also showed how AI agents can function on a computer interface.
With the market for agentic AI hot, some see Nvidia as taking advantage of and capitalizing on that buzz with the microservices in NeMo Guardrails.
An advantage for enterprises
The new NIM microservices benefit enterprises because they can add them to their agentic workflows, Cambrian AI Research analyst Karl Freund said.
"This is a big deal in that it helps Nvidia help their customers build agentic workflows with minimal programming," Freund said.
For enterprises, one of the biggest concerns that prevent them from deploying and scaling GenAI practice is a concern around safety, security and reliability, [and] responsibility.
Chirag DekateAnalyst, Gartner
The new tools are also beneficial because enterprises can run them in any environment, whether in the cloud or on-premises, Gartner analyst Chirag Dekate said.
"For enterprises, one of the biggest concerns that prevent them from deploying and scaling GenAI practice is a concern around safety, security and reliability, [and] responsibility," Dekate said.
The release of new NIM microservices in NeMo Guardrails is likely one of many that Nvidia will make in the coming months, especially as the vendor continues to reveal its strategy toward enterprises, Dekate continued.
"Nvidia has enterprise in its sight," he said, adding that while Nvidia has typically catered to cloud providers and hyperscalers, it's now strengthening its software portfolio to appeal to more enterprises.
To lock in or not
While enterprises seeking an easier entry point into AI agents might go with Nvidia, some in highly regulated industries that deal with litigation might need the AI hardware and software provider to do more to ensure agents do not go astray, Dekate said.
Some enterprises might also decide not to choose Nvidia because they want options, Freund said.
"The issue is people not wanting to feel locked in to Nvidia," he said.
Nvidia is among an early group of vendors offering governance features for agentic AI. Many enterprises that do not want to lag behind competitors will have little choice but to use agentic AI technology because it can unlock business value opportunities, said David Nicholson, an analyst at Futurum Group.
"Agents will offer the opportunity to do increasingly valuable things, which means increasingly dangerous things," he said. "They will be able to make increasingly risky decisions, so you absolutely have to trust them."
Nvidia is enabling enterprises to work with AI agents while providing a walled garden, or a sealed environment with a single vendor's products, Nicholson continued. Enterprises that want to remain competitive will have to deal with lock-in concerns about not only running on Nvidia's hardware, but also its software, he said.
Nvidia is letting enterprises test the effectiveness of the NeMo Guardrails NIM microservices by using Nvidia Garak, an open source tool for LLM and application vulnerability scanning.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
