# IBM and Red Hat contribute Praxis Policy Engine to the Praxis proxy

published: 2026-11-09T16:05:39.616Z
topic: identity
source: https://research.ibm.com/blog/building-policy-enforcement-into-the-ai-stack-with-praxis?utm_medium=rss&utm_source=rss
url: https://aiagentnews.io/stories/ibm-and-red-hat-contribute-praxis-policy-engine-to-the-praxis-proxy

IBM and Red Hat have contributed the Praxis Policy Engine (PPE) to Praxis, an open-source AI and cloud-native proxy server that has been onboarded into the CNCF Sandbox. PPE treats tools, prompts, resources, model calls and other agents as explicit policy targets.

IBM and Red Hat have contributed the Praxis Policy Engine to Praxis, an open-source AI and cloud-native proxy server. Praxis uses PPE as its policy filter for both ordinary and AI traffic, and IBM-affiliated contributors hold maintainer roles. The project has also been onboarded into the Cloud Native Computing Foundation Sandbox.

PPE combines identity, delegation, content controls and audit in one enforcement pipeline that runs outside the model's control. Teams can write hard access rules, softer behavior policies and domain-specific controls in a single policy specification. That specification can be deployed at a gateway, beside an agent, or inside a framework.

IBM Research says it is studying how host isolation and gateway policy enforcement could work together. A Praxis gateway could govern who can invoke an agent, which tools it can reach over the network and what data can leave its boundary. PPE can also emit an OCSF audit record for each policy decision, with optional signatures and hash chaining.

## Why it matters if you build agents

Developers building agents can enforce access, tool and data rules at a gateway or beside the agent instead of relying only on the model's behavior. Per-decision OCSF audit records give security teams a standard format to review what an agent was allowed to do.

The work is still research, and IBM says it plans to test whether agents and their tools can bypass these controls. Teams should treat the current code as a foundation to evaluate rather than a finished guarantee.