Nvidia is launching new NIM microservices within NeMo Guardrails to improve the security, accuracy and control of AI applications, specifically agents.
Nvidia has introduced new NIM microservices as part of NeMo Guardrails: a software collection for managing AI security. These microservices help companies implement generative AI solutions more securely and efficiently. They focus on preventing malicious or unwanted AI responses and provide protection against attempts to bypass restrictions.
The three new microservices include:
- Content safety microservice: prevents AI from generating inappropriate or harmful content.
- Topic control microservice: ensures conversations stay within allowed topics.
- Jailbreak detection microservice: helps prevent AI systems from being manipulated to provide unauthorized responses.
Optimization
The microservices use specialized low-latency language models and are optimized for industries such as healthcare, manufacturing and retail.
read also
Nvidia introduces new NIM microservices for AI security
Nvidia previously launched an AI Blueprint for retail, which uses the new microservices for improved customer interactions. To help developers test AI systems for vulnerabilities, Nvidia also offers Garak, an open-source toolkit for scanning AI applications for security issues such as prompt injections and data leaks.
In its announcement, Nvidia points specifically to agentic AI, in which AI thus acts semi-autonomously as an agent. In terms of security and guardrails, the distinction between agentic AI and an AI working exclusively with men is not so great. However, the greater degree of automy of an agent does make it all the more important that it function within expectations.
The NeMo Guardrails microservices and the Garak toolkit are now available to enterprises and developers.