Red Hat expands agentic AI strategy with new inference, automation and sovereignty capabilities

IBM Corp. subsidiary Red Hat today is unveiling a broad set of product and partnership announcements aimed at helping enterprises put artificial intelligence into operation, modernize infrastructure and extend open-source platforms into new environments ranging from software-defined vehicles to computing in space.

The announcements at Red Hat Summit in Atlanta extend Linux and container platforms into specialized environments and give enterprises greater operational control over hybrid cloud infrastructure. The company is also emphasizing new governance, sovereignty and security features as organizations move from experimentation to production AI deployments.

In a media briefing ahead of the event, Red Hat Chief Executive Matt Hicks described AI as a major technology inflection point comparable to Linux, open source and cloud computing. He argued that enterprises do not want to discard existing infrastructure investments in order to adopt AI.

“At Red Hat, we want our platforms to allow customers to embrace that balance upfront,” Hicks said. “That means in our platforms, allowing customers to use AI to its fullest extent while also improving what runs your business today.”

The centerpiece is Red Hat AI 3.4, an updated version of the company’s enterprise AI platform designed to support large-scale inferencing and agentic AI deployments across hybrid cloud environments.

The release introduces a new model-as-a-service capability designed to let enterprises expose internally approved AI models through governed interfaces while monitoring usage and applying policy controls.

Joe Fernandes, vice president and general manager of Red Hat AI, said the firm’s AI strategy is divided into four key pillars: scalable inference, connecting enterprise data to models and agents, managing agents across hybrid infrastructure and providing a unified AI platform spanning hardware and cloud environments.

The new model-as-a-service capability enables administrators to govern access to AI models through a centralized gateway, track usage and apply policies. Red Hat is also expanding support for distributed inferencing and introducing techniques such as speculative decoding to improve performance and reduce operating costs. Speculative decoding is a large language model inference optimization technique that accelerates text generation up to threefold without reducing output quality.

Fernandes said the company believes that inferencing, rather than model training, will become the dominant enterprise AI workload.

“What’s really going to drive inference demand exponentially is AI agents,” he said. “We provide a platform where customers can deploy and manage their AI agents across a hybrid infrastructure environment.”

Agent management

The company is also adding agent management and observability features, including tracing for inference calls and tool usage, as well as support for Model Context Protocol gateways and catalogs. Additional features include prompt management, automated evaluation tools and integrated AI safety testing capabilities powered in part by Red Hat’s recent acquisition of Chatterbox Labs Inc.

Red Hat also added prompt management tools that treat prompts as managed enterprise assets and introduced an evaluation hub designed to assess model and agent quality, safety and accuracy. The platform uses MLflow for experiment tracking and lifecycle management.

Fernandes said enterprises are focused less on building foundational models and more on operationalizing existing ones with proprietary enterprise data.

“Pretraining models from scratch is limited to a few very large organizations,” he said. “We find enterprise customers are more focused on consuming those models and then basically connecting them to their own data.”

A major focus of the release is AI inference efficiency. Red Hat said the platform now supports speculative decoding techniques in the vLLM inference server that can improve response speeds by two to three times while reducing inference costs.

The AI announcements also deepen Red Hat’s collaboration with Nvidia Corp. That includes support for Nvidia’s Blackwell architecture and upcoming Vera Rubin platform, as well as participation in Nvidia’s OpenShell project for AI agent sandboxing and secure execution.

The partnership also adds support for confidential containers running on Nvidia confidential computing infrastructure within Red Hat OpenShift sandboxed containers. Nvidia confidential computing is a hardware-based security framework that protects sensitive data, AI models, and applications. Red Hat said its approach is intended to protect AI workloads and agents even if other agents or systems are compromised.

Speed and stability

The company also announced several major Linux and infrastructure initiatives aimed at addressing diverging customer demands for both faster innovation cycles and longer infrastructure stability.

One of the most notable is Fedora Hummingbird Linux, a new rolling-release, image-based Linux distribution hosted within the Fedora community and designed specifically for AI-driven development environments.

Unlike traditional Linux distributions, Fedora Hummingbird Linux is designed for rapid upstream updates and “agent-native” software delivery pipelines with minimal security vulnerabilities.

The distribution is intended to simplify AI experimentation by removing registration barriers and supporting anonymous, automated deployment workflows for AI agents.

Red Hat also introduced Red Hat Hardened Images, a new catalog of minimal container images designed to support “zero-CVE” security strategies. Common Vulnerabilities and Exposures is a list of publicly disclosed cybersecurity flaws. The images contain only the components required for applications to run and include software bills of materials and cryptographic verification.

Hardened Images provide “a highly refined starting point for organizations that need to minimize their footprint without sacrificing the trust of the supply chain,” said Gunnar Hellekson, vice president and general manager of the Red Hat Enterprise Linux business.

Long-life Linux

The company also unveiled Red Hat Enterprise Linux Long-Life Add-On, which extends support for specific Red Hat Enterprise Linux releases indefinitely through annual renewals. The offering is aimed at industries such as aerospace, healthcare and telecommunications that operate infrastructure with multi-decade lifecycles.

“The market is simultaneously moving faster and moving slower,” Hellekson said. “We’re giving our customers ultimate control over their infrastructure timeline.”

On the developer side, Red Hat announced the general availability of Red Hat Desktop, a commercially supported version of the Red Hat build of the Podman Desktop open-source graphical tool that simplifies the use of containers, pods and Kubernetes on local developer machines.

The product includes isolated AI agent sandboxing capabilities intended to let developers safely test autonomous agents on local hardware without risking host operating systems.

With more than 4 million downloads, Podman Desktop is “the industry standard for working with Linux containers on your laptop, composing applications, iterating and deploying them then to anybody’s Kubernetes environment,” said Mike Barrett, vice president and general manager of Red Hat Hybrid Platforms.

The company also enhanced Red Hat Advanced Developer Suite with trusted software factory capabilities, trusted software libraries and AI-driven exploit intelligence designed to identify whether vulnerable code paths are reachable in runtime environments.

Orchestrated automation

On the automation front, Red Hat introduced Ansible Automation Platform 2.7 and a new automation orchestrator designed to coordinate deterministic, event-driven and AI-driven workflows through a single governance layer.

The automation orchestrator is intended to act as a trusted execution layer connecting AI-generated insights to production infrastructure actions.

Ansible 2.7 improves self-service automation so users in a branch office or on a factory floor can take control over servers and networks and update them on their own timetable, said Sathish Balakrishnan, vice president and general manager of Red Hat’s Ansible business unit.

New automation dashboards “give organizations a clear picture of the [return on investment] and financial impact that automation is delivering,” he said. “Customers are now able to use their own data to calculate savings and efficiencies.”

Balakrishnan emphasized that enterprises need consistent governance regardless of whether actions are triggered by humans, events or AI agents.

“Every action, whether it’s triggered by human, an event or an AI agent, passes through the same [access controls], approval gates, audit trails, content signing and credential management,” he said.

Red Hat is also expanding its support for sovereign clouds, which store and process data within specific national or regional borders to comply with local data privacy, security and legal regulations. It’s a market that Gartner Inc. estimates is growing 36% annually.

New compliance automation tools, isolated infrastructure deployment templates, on-premises telemetry capabilities and localized software delivery services are to help customers maintain regional control over infrastructure and operational data.

The company framed sovereignty less as a regulatory requirement and more as a strategic need for operational independence.

“Sovereignty is about control, where an organization can maintain oversight and command over its own trajectory regardless of geopolitical shifts, market dynamics or changing vendor terms,” the company said in its announcement.

Today’s announcements collectively reflect Red Hat’s broader effort to position open-source infrastructure as the operational backbone for enterprise AI deployment as enterprises seek flexibility among models and agents.

“Open source has shown to be the best innovation model at a global scale,” Hicks said. “And we believe that AI will only amplify that capability.”

Image: SiliconANGLE/Microsoft Designer