AgentOps tools now cover the full AI lifecycle from development to production, adding tracing, observability, identity controls, and lifecycle governance to address safety and governance concerns for autonomous systems.

The rollout supports NVIDIA Blackwell GPUs and AMD MI325X at launch and will run in managed cloud environments, including IBM Cloud, to standardize operations across hardware and providers.

Inference improves with Red Hat AI Inference, featuring request prioritisation to balance interactive and background traffic, plus speculative decoding to speed responses while reducing costs.

Top executives and partners stress openness and enterprise standardization, highlighting collaboration with CoreWeave and NVIDIA to enable scalable, secure autonomous systems across environments.

Red Hat AI 3.4 debuts as a platform centered on a unified framework for running models and autonomous agents across hybrid clouds, moving AI from pilots to governed, scalable operations.

A Model-as-a-Service interface lets developers access approved models while admins monitor usage, enforce policies, and track behavior via identity-based authentication and consumption metrics.

Governance and auditability are core, using cryptographic identity management with SPIFFE/SPIRE to replace static keys with short-lived tokens for least-privilege access and traceability of agent actions to verified identities.

Inference support extends beyond OpenShift to CoreWeave and Azure Kubernetes Service, ensuring consistent inference environments across on-premises and cloud deployments.

MLflow integration enables experiment tracking and artifact management for generative and predictive AI, with visibility into agent execution, including language model calls, reasoning steps, tool usage, responses, and token usage.

A prompt management system and evaluation hub govern prompts as data assets, with tools to assess quality, safety, and risk across models, apps, and agents.

Automated safety testing and adversarial scanning are being built in, leveraging Chatterbox Labs and Garak for jailbreak, prompt injection, and bias risks, with NVIDIA NeMo Guardrails providing runtime safety controls.