NVIDIA Announces NIM Microservices to Protect Agentic AI Applications. To improve accuracy, security, and control for businesses developing AI across industries, NVIDIA NeMo Guardrails incorporates new NVIDIA NIM microservices.
With “knowledge robots” that can perform a range of jobs, artificial intelligence (AI) agents have the potential to revolutionise productivity for the billions of knowledge workers worldwide. Businesses must handle important issues including trust, safety, security, and compliance in order to develop AI agents.
The NVIDIA NeMo Guardrails software toolkit includes new NVIDIA NIM microservices for AI guardrails, which are portable, optimized inference microservices that assist businesses in enhancing the security, accuracy, and scalability of their generative AI systems.
NeMo Guardrails, a component of the NVIDIA NeMo platform for curating, customising, and guardrailing AI, is essential to the microservices’ orchestration. In large language model (LLM) applications, NeMo Guardrails assists developers with the integration and management of AI guardrails. NeMo Guardrails is being used by industry heavyweights Amdocs, Cerence AI, and Lowe’s to protect AI applications.
The NIM microservices can be used by developers to create more reliable and secure AI agents that are strengthened against jailbreak attempts and offer suitable, safe answers under context-specific guidelines. Agents can increase client happiness and trust when used in customer service across a variety of industries, including manufacturing, retail, healthcare, banking, and automobiles.
The Aegis Content Safety Dataset, one of the best human-annotated data sources in its area, was used to train one of the new microservices designed for content safety moderation. NVIDIA owns and curates the dataset, which is openly accessible on Hugging Face and contains more than 35,000 human-annotated data samples that have been reported for AI safety and efforts to circumvent system limitations through jailbreaks.
NVIDIA NeMo Guardrails Keeps AI Agents on Track
AI is quickly increasing the efficiency of many corporate operations. It helps resolve client concerns up to 40% faster in customer service. Scaling AI for customer service and other AI agents, however, necessitates safe models that guarantee the AI application operates within predetermined bounds and avoid damaging or unsuitable outputs.
For NeMo Guardrails, NVIDIA has unveiled three new NIM microservices that enable AI agents to function at scale while exhibiting regulated behaviour:
- The content safety NIM microservice ensures that replies adhere to ethical norms by protecting AI from producing biased or damaging outputs.
- Topic control NIM microservice that prevents unwanted content or digression and keeps discussions on authorized topics.
- The NIM microservice for jailbreak detection helps preserve AI integrity in hostile situations by providing defense against jailbreak attempts.
Because a one-size-fits-all approach fails to adequately safeguard and govern complex agentic AI workflows, developers can fill in the gaps that may arise when only more broad global regulations and protections are in place by implementing numerous lightweight, specialised models as guardrails.
Small language models, such as those included in the NVIDIA NeMo Guardrails collection, have reduced latency and are made to function well even in distributed or resource-constrained settings. This makes them perfect for scaling AI applications in settings like hospitals or warehouses in sectors like manufacturing, healthcare, and automotive.
NeMo Guardrails Help Industry Leaders and Partners Protect AI
NeMo Guardrails, an open-source tool, assists developers in coordinating several AI software policies, or rails, to improve the security and control of LLM applications. It provides a strong platform for creating AI systems that can be scaled without sacrificing performance or safety by integrating with NVIDIA NIM microservices.
NVIDIA NeMo Guardrails is being used by Amdocs, a top global supplier of software and services to media and communications firms, to improve AI-driven client interactions by providing safer, more precise, and contextually relevant responses.
NeMo Guardrails and other technologies are crucial for protecting generative AI systems and ensuring their ethical and safe operation. This are strengthening the amAIz platform’s ‘Trusted AI’ capabilities to provide scalable, safe, and dependable agentic experiences by using NVIDIA NeMo Guardrails. This establishes new benchmarks for AI innovation and operational excellence by enabling service providers to confidently and safely implement AI solutions.
With the help of NVIDIA NeMo Guardrails, Cerence AI, a business that specializes in AI solutions for the automotive sector, is making sure that its in-car assistants provide safe, contextually relevant interactions using its CaLLM family of large and small language models.
NeMo Guardrails enables us to provide the automobile clients with dependable, context-aware solutions as well as logical, considerate, and hallucination-free answers. To CaLLM family of language models is protected from delivering unwanted or improper content to end users by NeMo Guardrails, which is also customizable for it automobile customers and aids in filtering damaging or disagreeable requests.
Leading home improvement company Lowe’s is utilising generative AI to expand on the extensive knowledge of its employees. By giving associates better access to thorough product knowledge, these technologies enable them to respond to enquiries from customers, assisting them in locating the appropriate products to finish projects and establishing new benchmarks for customer satisfaction and retail innovation.
At the NRF show, NVIDIA recently revealed that its NVIDIA AI Blueprint for retail shopping assistants integrates NVIDIA NeMo Guardrails microservices to create more dependable and controlled customer interactions during digital shopping experiences, further speeding up the adoption of AI safeguards in AI application development and deployment in retail.
NeMo Guardrails is also being incorporated into the solutions of consulting giants Taskus, Tech Mahindra, and Wipro to give their enterprise clients generative AI applications that are safer, more dependable, and easier to control.
NeMo Guardrails is open and extendable, providing AI observability and development tools along with integration with a strong ecosystem of top AI safety model and guardrail suppliers. It facilitates interaction with ActiveFence’s ActiveScore, which offers visibility, analytics, and monitoring while filtering offensive or dangerous content in conversational AI applications.
NVIDIA NeMo Guardrails makes it simple to integrate and orchestrate Hive’s AI-generated content detection models for audio, video, and image content as NIM microservices in AI applications.
NeMo Guardrails and the Fiddler AI Observability platform may be seamlessly integrated to improve AI guardrail monitoring capabilities. Additionally, W&B Weave’s functionality is being enhanced by Weights & Biases, an end-to-end AI development platform, through interactions with NeMo Guardrails microservices. Weights & Biases’ current portfolio of NIM integrations for optimal AI inferencing in production is expanded upon by this improvement.
NeMo Guardrails Provides Open-Source AI Safety Testing Tools
NVIDIA Garak is an open-source toolkit for LLM and application vulnerability assessment created by the NVIDIA Research team for developers who are prepared to evaluate the efficacy of implementing safeguard models and other measures.
Developers can use Garak to find vulnerabilities in systems by evaluating them for problems including data leaks, prompt injections, code hallucinations, and jailbreak scenarios employing LLMs. Garak assists developers in identifying and fixing possible flaws in AI models to improve their robustness and safety by producing test cases with improper or inaccurate results.
Accessibility
Developers and businesses may now access the NVIDIA Garak toolkit, NeMo Guardrails microservices, and NVIDIA NeMo Guardrails for rail orchestration. This article demonstrates how developers may use NeMo Guardrails to begin integrating AI safeguards into AI customer service agents.