Red Hat Launches “Red Hat AI 3.5,” Bolstering Enterprise AI Safety and Observability

Red Hat announced major updates across its AI portfolio on the 10th and released “Red Hat AI 3.5.” The company said enterprise AI usage is moving beyond early experiments and pilot projects into full-scale deployment, and that IT and platform engineering teams now need to manage AI at the same level required for mission-critical infrastructure.

Against that backdrop, Red Hat introduced the new version to improve the safety of enterprise AI operations and enhance management efficiency. The company said Red Hat AI 3.5 is focused on providing a foundation for control, security, and observability for AI workloads across the hybrid cloud.

Red Hat AI 3.5 integrates safety validation, multitenancy, agentic development, and observability capabilities. Red Hat said the goal is to reduce problems caused by fragmented tools and ease operations that rely heavily on manual work.

Red Hat AI 3.5 expanded its capabilities by adding safety and observability features. On the safety side, EvalHub supports validation before model deployment, risk-focused safety benchmarking, and the generation of compliance certificates. On the observability side, the new observability dashboard provides real-time metrics for inference status, GPU utilization, and AI model performance. In addition, even users without administrator privileges can check per-user token usage and access the showback dashboard and the distributed inference workload dashboard.

Red Hat AI 3.5 also expands enterprise platform capabilities. To address AI use cases that require complete isolation, it improved multitenancy features, targeting environments that require full isolation across the entire hardware-to-software stack, including AI service providers. It also supports native multitenancy in shared GPU environments and supports priority-aware serving.

Red Hat officially supports Red Hat OpenShift Hosted Control Planes operations in Red Hat OpenShift Virtualization deployment environments. It also adds VM-level isolation across shared GPU infrastructure when running AI workloads on Red Hat OpenShift Virtualization VMs.

With this setup, infrastructure providers can operate and upgrade the entire underlying environment from a single control point.

Red Hat AI 3.5 supports building AI agents with governance applied. AutoRAG directly connects enterprise data repositories with agentic applications. AutoRAG supports multilingual documents and also provides conversational testing and context-based search capabilities.

Red Hat AI 3.5 enables users to review and validate RAG configurations before deployment through visualized pipelines. It also provides preconfigured implementations of agent templates, with target patterns including enterprise use cases such as code review, document processing, and research workflows. After deployment, it applies “Inference-Time Scaling.”

This feature uses “Inference-Time Scaling” to adjust compute resources based on query complexity, with the goal of managing GPU costs.

Joe Fernandes, vice president and general manager of Red Hat’s AI business unit, said the industry is moving from the stage of introducing AI into production environments centered on discussion to a stage of large-scale operations built on trustworthy enterprise infrastructure. He said the requirements for large-scale operations include safety validation materials, governed agents, cost allocation, and multitenancy. He added that Red Hat AI 3.5 provides operational control, verifiable reliability, and an agentic foundation for AI operations across the hybrid cloud, and described the operational goal as a safe, controlled, and accountable enterprise AI architecture.

Source: IT DAILY · Yang Seung-gap
Original: https://www.itdaily.kr/news/articleView.html?idxno=241523

References

• https://www.itdaily.kr/news/articleView.html?idxno=241523