IBM adds an open-source policy engine to the Praxis AI gateway
The engine checks each tool, prompt and model call against one set of rules that the model cannot change.
IBM and Red Hat gave the Praxis Policy Engine to Praxis, an open-source proxy server (software that sends traffic between users and AI systems). Praxis uses the engine as its policy filter for AI traffic and other traffic. Praxis is at this time in the Sandbox of the Cloud Native Computing Foundation. IBM gave no test results, dates or prices.
How it works
The rules include tools, prompts, resources, model calls and other agents. One policy specification holds hard access rules, behavior rules and rules for each domain. Teams can put it at a gateway, adjacent to an agent or in a framework.
The proof
IBM gave no numbers. It says the engine works in Rosso Cortex for its laptop policy build. The engine can write an audit record of each decision, with optional signatures that show a change to the record.
What is not known
IBM says it will test how agents and tools can bypass these controls. IBM gives no data on which controls stop a bypass. IBM also works with CoSAI on a standard for agent identity.
Insight: IBM gives away the rule layer and sells the gateway
On 27 October 2026, IBM added an AI Gateway to watsonx Orchestrate for agents from other firms. At this time IBM puts its policy engine in open-source Praxis. On 21 October 2026, Solcis released an engine for agent rules. This reads as a contest to set the standard for agent rules, because open code in a foundation can become the default. Check which format your vendor uses.
Sources
- Building policy enforcement into the AI stack with Praxisresearch.ibm.com
Posted