Kubernetes AI Conformance

A standardized approach to running AI/ML workloads on Kubernetes

Get Certified · Contribute · FAQ · AI Conformance Project

If you're here to get certified, start at For Vendors. If you're here to help shape the program, jump to For Contributors.

• What is this?
• Getting Started
  • For Vendors
  • For Contributors
• How Certification Works
• Requirements Overview
• Resources
• Community

The Kubernetes AI Conformance Program defines the capabilities a Kubernetes platform needs to reliably run AI and machine learning workloads. The goal is simple: if your AI application works on one conformant platform, it should work on others too—with fewer "it works on my cluster" surprises.

AI/ML workloads tend to stress clusters in unique ways (accelerators, bursty traffic, and strict isolation). Today, those capabilities vary across platforms. This program aims to:

• Make AI/ML workloads more portable across Kubernetes platforms
• Reduce platform-specific workarounds and "it works on my cluster" surprises
• Give the AI tooling ecosystem a clear baseline to build and test against

We're focusing on the most common AI/ML use cases:

• Training - Distributed or large training jobs that need accelerators and predictable scheduling
• Inference - Model/LLM serving where latency, routing, and scaling matter
• Agentic workloads - Multi-step workflows that combine tools, memory, and long-running tasks

If you provide a Kubernetes platform and want to get certified, here's what you need to know.

Most submissions are a completed checklist plus links to public evidence—think of it as a structured, reviewable self-assessment.

Before you start: Your platform must already be Kubernetes Conformant. AI conformance builds on top of base Kubernetes conformance.

For the canonical reference tests and example implementations, see the upstream AI Conformance test suite.

Starting with v1.37, we support a hybrid verification approach that combines automated test results with manual attestation:

1. Automated Tests: For requirements with automated tests (e.g., Secure Accelerator Access, Gang Scheduling), it is recommended that you run the upstream AI Conformance test suite and include the generated test artifacts (e2e.log, junit.xml, or results.json) in your submission PR.
2. Manual Attestation: For requirements without automated tests, continue to provide documentation or reference URLs in the evidence field of the checklist YAML.

Reference your automated test results in the checklist YAML using relative file paths or the file:// scheme. Starting with v1.37, it is recommended that any requirement covered by an automated test is verified using this test suite.

1. Prepare - Review the certification requirements and make sure your platform meets them
2. Document & Test - Fill out the conformance checklist, run automated tests if applicable, and gather evidence (documentation, test results, etc.)
3. Submit - Create a pull request with your submission to the https://github.com/cncf/k8s-ai-conformance repo.
4. Review - CNCF reviews your submission (typically takes up to 10 business days)

• A completed conformance checklist (YAML file)
• Public documentation showing how your platform meets each requirement
• (Recommended for v1.37+) Automated test artifacts (e2e.log, junit.xml, or results.json) for requirements covered by automated tests
• Your product logo in vector format (SVG, EPS, or AI)
• Proof of Kubernetes conformance

Note: Starting with v1.37, automated tests are available and recommended for applicable requirements. For earlier versions or requirements without automated tests, certification is based on documentation-backed self-assessment.

For detailed instructions on running tests and what to include, see instructions.md.

The program is a community-led effort to establish a vendor-neutral baseline for AI portability. We welcome participation from all stakeholders, including end users, to ensure the standard remains independent and effective.

1. Open an issue — have an idea, question, or suggestion? Start here.
2. Join the AI Conformance project meetings.
3. Check out the project boards to see what's in progress.

flowchart TD
    A[Platform must be Kubernetes Conformant] --> B[Complete checklist & run automated tests]
    B --> C[Gather evidence and test artifacts]
    C --> D[Submit pull request]
    D --> E[CNCF reviews submission]
    E -->|Approved| F[Certified for 1 year]
    E -->|Needs changes| B

Important notes:

• Certifications are valid for one year and must be renewed
• Certification is per-product and per-configuration (e.g., cloud vs air-gapped)
• The conformance requirements are aligned with Kubernetes release cycles

Platforms need to demonstrate capabilities across several areas: accelerators, networking, scheduling, observability, security, and operator support. The specifics evolve with each Kubernetes release.

For the full and up-to-date requirements, see the conformance versions in the WG repo.

• FAQ - Common questions about the program
• Instructions - How to prepare and submit your conformance results
• Terms & Conditions - Legal requirements for certification

Pick the one that matches your Kubernetes version:

• AIConformance-1.37.yaml (latest)
• AIConformance-1.36.yaml
• AIConformance-1.35.yaml

• Kubernetes Conformance - Base Kubernetes certification (required first)
• AI Conformance Tests - Canonical reference tests for AI workload validation
• Conformance Tests - Kubernetes test suite
• Verify Conformance Bot - Automation coming soon

The Kubernetes AI Conformance project governs this program and defines the conformance requirements.

• Meetings: Check the AI Conformance project meetings
• Contact: sig-architecture@kubernetes.io (mailing list) and Slack channel

See all certified platforms in the version directories:

• v1.36/
• v1.35/
• v1.34/
• v1.33/

• Read the FAQ first
• Ask questions in the Kubernetes AI Conformance project
• Open an issue
• Email conformance@cncf.io

For private review of unreleased products, contact conformance@cncf.io directly.

Apache License 2.0 - see LICENSE for details.