OpenShift 5 and RHACS 5 tighten controls around AI workloads and vulnerability triage

Red Hat has detailed a security package around OpenShift 5 that spans workload isolation, AI-agent governance, vulnerability prioritization and software-supply-chain controls. The announcement also introduces Red Hat Advanced Cluster Security for Kubernetes 5, or RHACS 5, with changes intended to help security teams distinguish urgent findings from background vulnerability noise (Red Hat).

Protecting AI workloads in use

OpenShift 5 applies network policies during installation to restrict traffic into and out of control-plane operand namespaces. It also advances Red Hat’s post-quantum work through hybrid key exchange for compatible TLS 1.3 clients (Red Hat).

For AI workloads, Red Hat says confidential containers can encrypt model weights in CPU and GPU memory. Confidential GPU support is generally available on supported bare-metal systems, while the Red Hat build of Agent Sandbox remains a Technology Preview for isolating untrusted agent-generated code in virtual machines (Red Hat).

The OpenShift MCP server connects agent access to identity integration and Kubernetes role-based access controls, with audit trails for reviewing activity. Red Hat Connectivity Link’s MCP gateway can also use NVIDIA NeMo Guardrails to screen tool requests and responses for prompt injection and sensitive-data exposure (Red Hat).

Adding context to vulnerability work

RHACS 5 incorporates the US Cybersecurity and Infrastructure Security Agency’s Known Exploited Vulnerabilities catalog into vulnerability prioritization. Teams can use that exploitation evidence in build and deployment policies rather than treating every finding as equally urgent (Red Hat).

A new CVE Origin indicator distinguishes Red Hat assessments from OSV.dev assessments. RHACS 5 also makes vulnerability management for supported RHEL virtual machines generally available and adds scheduled compliance profiles for OpenShift Virtualization (Red Hat).

Red Hat is separately introducing an RHACS MCP server as a Technology Preview. It provides read-only CVE lookup through compatible assistants, with results limited to the scope of the RHACS API token used by the agent (Red Hat).

Extending trust into delivery

The same package reaches the supply chain. Trusted Profile Analyzer 3.1 automatically flags machine-learning and cryptographic components in software bills of materials, while Trusted Artifact Signer 1.5 adds AI-model signing and verification plus transparency-log search through its console (Red Hat).

For platform teams, the practical change is breadth: OpenShift 5’s security story is not a single control but a chain running from cluster network boundaries and confidential execution through agent authorization, vulnerability triage and model provenance. Teams evaluating the release should still check the support status of each component, because the package mixes generally available features with Technology Preview capabilities (Red Hat).