The Open Group’s Zero Trust & AI Security Extensions are a set of emerging standards that extend the classic Zero Trust model into AI systems—covering AI agents, data pipelines, model access, and governance. They build on The Open Group’s authoritative Zero Trust Body of Knowledge and integrate with global standards such as NIST, MITRE, OWASP, ISO/IEC, and the EU AI Act.

Below is a clear, structured explanation of what these extensions are and how they fit into the broader Zero Trust ecosystem.

What The Open Group’s Zero Trust & AI Security Extensions Are

The Open Group is developing a modular, standards‑based security body of knowledge that includes:

• Zero Trust Commandments

• Zero Trust Reference Model

• Security Principles for Architecture

• Security Matrix (in development)

• Enterprise Risk Integration (planned)

These components form the foundation for extending Zero Trust into AI systems, especially as AI introduces new trust boundaries and attack surfaces.

Although The Open Group has not yet released a standalone “AI Security Standard,” the Zero Trust Reference Model and Commandments are already being extended to AI through:

• AI‑specific trust boundaries

• AI agent identity and privilege models

• Data governance for training and inference

• Integration with OWASP AI Exchange and EU AI Act security controls

• Alignment with Microsoft’s Zero Trust for AI (ZT4AI) guidance

Core Components Relevant to AI Security

1. Zero Trust Commandments (Published Standard)

These are prescriptive “must/shall” statements that define how to apply Zero Trust across any system—including AI.
They emphasise:

• Identity‑first security

• Least privilege

• Continuous verification

• Strong separation of duties

• Secure-by-design architecture

These principles map directly to AI workloads, where models, agents, prompts, plugins, and data sources must all be treated as separate identities with explicit trust boundaries.

2. Zero Trust Reference Model (Snapshot Release)

This model provides the architectural blueprint for implementing Zero Trust across:

• Identity

• Devices

• Networks

• Applications

• Data

• Workloads

AI systems fit into this model by treating:

• Models as workloads

• Agents as identities

• Prompts as untrusted input

• Plugins/tools as external services

This aligns with Microsoft’s Zero Trust for AI approach, which explicitly extends Zero Trust to the full AI lifecycle.

3. Security Principles for Architecture

These principles help architects embed consistent security across all systems—including AI pipelines.
They support:

• Secure data ingestion

• Secure model training

• Secure inference

• Secure agent behaviour

• Governance and auditability

These principles are foundational for AI security reference architectures.

4. Security Matrix (In Development)

The Security Matrix will map:

• Controls

• Roles

• Responsibilities

• Threats

• Mitigations

…across traditional and AI‑enabled systems.
This will become a key tool for AI governance and risk management.

How These Extensions Apply to AI

The Open Group’s Zero Trust extensions support AI security by:

A. Defining AI Trust Boundaries

• User → Agent

• Agent → Model

• Model → Data

• Agent → Tools/APIs

• Model → Vector Stores

B. Applying Zero Trust Principles to AI

• Verify explicitly: authenticate agents, models, plugins, and data sources

• Least privilege: restrict model access, tool use, and agent permissions

• Assume breach: design for prompt injection, data poisoning, and model misuse

C. Integrating with Global Standards

The Open Group’s work aligns with:

• NIST AI RMF

• EU AI Act

• OWASP AI Exchange

• MITRE ATLAS

• Microsoft ZT4AI

This makes it a neutral, standards‑aligned foundation for AI security architectures.

Summary Table

Component - Purpose AI Relevance

Zero Trust Commandments

Prescriptive rules for Zero Trust  - Applies to agents, models, data, plugins

Zero Trust Reference Model

Architectural blueprint - Defines AI trust boundaries

Security Principles for Architecture

Embeds security into design - Supports secure AI pipelines

Security Matrix (future)

Controls & roles mapping - Will map AI threats & mitigations

Enterprise Risk Integration (future)

Risk quantification - Supports AI risk governance