What Is Machine Genome?
Machine Genome is an open identity and provenance protocol for the entire AI artifact ecosystem — models, agents, harnesses, datasets, and the connections between them. Released July 20, 2026 by paxlabs-inc, it's already garnered 269 stars.
Why does this matter? As AI agents become autonomous — negotiating APIs, sharing data, executing tasks — verifiable identity is critical. Without it: Who built this model? What data was it trained on? Has this agent been tampered with?
Machine Genome provides a standardized, machine-readable format — a genome for AI artifacts — encoding identity, lineage, dependencies, and attestations in a cryptographically verifiable package. Apache 2.0 licensed, interoperable, and built for the multi-agent future.
Key Features
Identity Protocol for AI Artifacts
Every model, agent, dataset, and harness gets a unique, cryptographically signed identity document.
Provenance Tracking
Full lineage: training data → model weights → fine-tuned variants → deployed agents. Every step is recorded.
Cross-Platform Compatibility
Works across Hugging Face, Ollama, local infrastructure, and cloud providers. No vendor lock-in.
Open Standard (Apache 2.0)
Fully open specification and reference implementation. Free for commercial and non-commercial use.
Cryptographic Attestations
Signatures, hashes, and verifiable credentials ensure artifacts haven't been tampered with.
Dependency Declaration
Explicitly declares what other artifacts an agent or model depends on, with version pinning.
Comparison: Machine Genome vs. Existing Standards
How does Machine Genome stack up against existing standards? Here's a feature-by-feature comparison.
| Feature | Machine Genome | Model Cards | Datasheets for Datasets | SBOM | W3C Verifiable Credentials | Hugging Face Metadata |
|---|---|---|---|---|---|---|
| Machine-readable format | ✔ | ~ | ~ | ✔ | ✔ | ✔ |
| Cryptographic signatures | ✔ | ✘ | ✘ | ~ | ✔ | ✘ |
| AI-specific fields | ✔ | ✔ | ✔ | ✘ | ✘ | ✔ |
| Dependency tracking | ✔ | ✘ | ✘ | ✔ | ~ | ~ |
| Agent identity support | ✔ | ✘ | ✘ | ✘ | ✔ | ✘ |
| Provenance lineage | ✔ | ~ | ✔ | ✔ | ~ | ~ |
| Cross-platform | ✔ | ✘ | ✘ | ~ | ✔ | ✘ |
| Open standard (license) | ✔ | ✔ | ✔ | ✔ | ✔ | ~ |
✔ = Full support · ~ = Partial support · ✘ = Not supported
How Each Standard Stacks Up
📋 Model Cards (Google)
Model Cards are documentation templates — human-readable reports for model transparency. They lack machine-verifiable identity, cryptographic signatures, and dependency tracking. Machine Genome complements them with a machine-readable identity layer.
📊 Datasheets for Datasets (Microsoft)
Datasheets for Datasets pioneered dataset provenance documentation. They're strong for dataset docs but offer no agent identity, cross-platform portability, or signed attestations. Machine Genome shares the provenance philosophy while adding verifiability and broader coverage.
📦 SBOM (Software Bill of Materials)
SBOM is the gold standard for software dependency tracking (SPDX, CycloneDX). Machine Genome borrows its dependency graph concept but extends to AI-specific artifacts. The two are interoperable — an AI system can have both.
🪪 W3C Verifiable Credentials
W3C Verifiable Credentials provide a general-purpose framework for verifiable identity claims. Machine Genome builds on VC concepts but adds AI-specific semantics: model architecture hashes, training lineage, capability declarations, and policy bindings. A Machine Genome identity can be wrapped in a W3C VC.
🤗 Hugging Face Model Hub Metadata
Hugging Face's metadata (YAML headers) provides structured fields for tags, pipeline types, and licenses — but is tightly coupled to their ecosystem. Machine Genome is platform-agnostic with cryptographic verification, dependency declarations, and multi-hop provenance not natively supported in Hugging Face metadata.
Protocol Decision Flowchart
Not sure which approach is right for your AI project? Follow this decision tree.
Identity + provenance + crypto
Documentation-only approach
Provenance + intended use
Adds crypto + lineage
Use SPDX or CycloneDX
Both for full coverage
Wrap MG in VCs for interop
Use Cases
AI Agent Identity Verification
When autonomous agents negotiate and interact, each presents its genome — a cryptographically signed identity proving its origin, capabilities, and policy constraints. Trust between agents becomes verifiable, not assumed.
Dataset Provenance Tracking
Trace every dataset to its source — collection methodology, preprocessing, license, and derivative datasets. Audit-ready.
Model Lineage Tracking
From base model through fine-tuning, quantization, and distillation — every transformation recorded with parent references.
Compliance & Auditing
Machine Genome provides the audit trail: who built what, when, with which data, under what policies. Exportable, verifiable.
Implementation Checklist
Ready to adopt an identity protocol for your AI artifacts? Here's your step-by-step checklist.
- Assess your needs: Which artifacts need identity — models, agents, datasets?
- Review the spec: Read the Machine Genome specification.
- Choose your tooling: Install the reference implementation (Python, Go, TypeScript).
- Generate your first genome: Create a genome for your most critical model or dataset.
- Integrate signing: Set up cryptographic signing with your existing key management.
- Publish genomes: Alongside your artifacts on Hugging Face, your registry, or API.
- Verify lineage: Confirm downstream artifacts reference their parent genomes.
- Set up agent verification: Configure genome verification on agent startup.
- Document policies: Attach usage terms & safety constraints to agent genomes.
- Audit & iterate: Run periodic audits. Update genomes as artifacts evolve.
Frequently Asked Questions
Tools & Infrastructure for AI Projects
Build and monitor your AI infrastructure with these recommended tools.
GPU Droplets for training and serving AI models. Simple, predictable pricing.
Get Started →Estimate AI infrastructure costs. Compare providers & optimize spending.
Calculate →