AI Provenance Protocol (APP)
The open standard for AI-generated content provenance.
APP is a vendor-neutral, machine-readable format for recording, embedding, and verifying the provenance of AI-generated content. It answers the fundamental questions about any AI output: who triggered it, what model produced it, what inputs were used, and whether a human reviewed it.
Get Started | Read the Specification | GitHub
Why APP?
AI-generated content is everywhere — product descriptions, legal documents, marketing copy, code, medical summaries. Yet there is no standard way to communicate its provenance.
The EU AI Act Article 50 (enforceable 2 August 2026) requires AI providers to mark output in a machine-readable format and offer free detection mechanisms. APP satisfies these obligations by design.
What APP provides
| Metadata schema | A JSON object (_ai_provenance) with five required fields — simple enough to implement in minutes |
| Embedding modes | Inline JSON, HTTP headers, or linked metadata — works with any content type |
| Verification protocol | Public REST endpoints for third-party provenance confirmation |
| Human review tracking | Record who reviewed AI output and when — critical for EU AI Act exemptions |
| Extension system | Domain-specific metadata via namespaced extensions (C2PA, MCP, and more) |
Quick example
{
"title": "Premium Leather Wallet",
"description": "Handcrafted from full-grain leather...",
"_ai_provenance": {
"app_version": "1.0.0",
"ai_generated": true,
"generator": {
"platform": "my-platform",
"model": "anthropic/claude-sonnet-4"
},
"generated_at": "2026-03-06T14:22:00Z",
"generation_id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
"verification_uri": "https://verify.my-platform.com/v1/verify",
"review": {
"human_reviewed": false
}
}
}Any system receiving this JSON instantly knows the content was AI-generated, which model produced it, and where to verify the claim.
Designed for the ecosystem
APP complements existing standards — it doesn’t replace them.
| Standard | Relationship |
|---|---|
| C2PA / Content Credentials | C2PA handles images, video, and audio. APP handles text and structured data. |
| Model Context Protocol (MCP) | MCP connects AI to tools. APP records what the AI produced. |
| Schema.org | APP metadata coexists with Schema.org properties. |
| W3C PROV | APP maps to the W3C PROV data model. |
EU AI Act ready
| Requirement | APP Feature |
|---|---|
| Machine-readable AI marking | ai_generated: true in _ai_provenance metadata |
| Detectable as AI-generated | Three embedding modes |
| Free detection mechanism | Public verification endpoints |
| Human review exemption | review.human_reviewed with audit trail |
| Non-breaking marking | Metadata is ignorable by non-APP systems |
Open and free
- Specification: CC BY 4.0
- SDKs and tools: Apache 2.0
- No proprietary claims: Implement APP without permission or fees
- Community governed: Open development on GitHub
Originally created at merchi.ai and maintained by the open-source community.