Introducing Enterprise Technologies in Asia
Subscribe

NVIDIA connects its NeMo Guardrails collection with NIM microservices to streamline agentic AI development

NVIDIA connects its NeMo Guardrails collection with NIM microservices to streamline agentic AI development

NVIDIA has announced they are enhancing the NeMo Guardrails feature with NIM microservices to allow enterprises to train both public and private models that adheres to safety and security while complying to various responsible rules and trust factors.

One of the new highlights of the update comes in the form of a content safety tool trained on the Aegis Content Safety Dataset, a high-quality, human-annotated data source curated by NVIDIA and publicly available on Hugging Face. With over 35,000 flagged samples, this dataset bolsters AI safety and protects against attempts to bypass system restrictions.

With it, LLMs will become more resilient against jailbreak attempts while delivering context-appropriate and safe responses toward mission-critical operations in sectors such as automotive, healthcare, retail, and finance.

In terms of actual NIM microservices, there's 3 which are:

  1. Content Safety Microservice: Ensures AI-generated outputs are free from bias and align with ethical standards.
  2. Topic Control Microservice: Keeps conversations focused on approved topics, preventing digressions or inappropriate content.
  3. Jailbreak Detection Microservice: Detects and protects against attempts to compromise AI integrity in adversarial scenarios.

These lightweight, specialized models address gaps in broader AI policies and are designed for efficient deployment in resource-constrained environments such as hospitals or warehouses.

Industry Adoption

Industry leaders like Amdocs, Cerence AI, and Lowe’s have already adopted NeMo Gaurdrails into AI systems with Amdocs strengthening its amAIz platform to ensure AI-driven customer interactions are safe and contextually appropriate, Cerence AI enhancing its in-car assistants with context-aware, secure solutions, ensuring accurate and ethical interactions, and Lowe's empowering store associates with AI tools for safer and more reliable customer interactions, boosting retail innovation and satisfaction.

Consulting firms like Taskus, Tech Mahindra, and Wipro are also incorporating NeMo Guardrails into their enterprise solutions, enabling safer generative AI applications for their clients. The framework integrates with leading AI observability tools, such as ActiveFence, Fiddler AI, and Weights & Biases, to further enhance AI safety and monitoring capabilities.

Lastly, for developers looking for some sweet and easy-to-use tools you may try out NVIDIA Garak which is an open-source toolkit for LLM vulnerability scanning that can identify potential issues such as data leaks, prompt injections, and jailbreak scenarios, helping developers address weaknesses and strengthen their AI systems.

Availability

NVIDIA NeMo Guardrails microservices, including the Garak toolkit, are now available to developers and enterprises. These tools provide a solid foundation for building safe and scalable AI systems, with tutorials to help developers integrate safeguards into their applications.