aidecl

Disclose your project's AI usage in one file.

aidecl.yaml is a small file in your repository. It lists the AI tools that were used, what they were used for and how much of the work they produced, or it states that no AI was used. A named person or team is responsible for it, and your CI validates it like any other file.

pip install aidecl
zsh
 
aidecl.yamlnew
schema_version: "1.0.0"project:name: pySDCcontent_type: softwareai_usage:used: truelevel: moderatecode_proportion:ai_generated_percent: 19ai_assisted_percent: 2human_only_percent: 79declaration:date: "2026-09-28"declared_by: Robert Speck
aidecl / validatequeued

0% written by hand0% AI-assisted0% AI-generated

These figures come from the declaration published by the pySDC project.

Adopted by

The Netherlands eScience Center's Generative AI Use Policy names the AI Declaration Format as a way to disclose AI use in publicly shared software, publications and datasets.

The pySDC project also publishes an aidecl.yaml in its repository.

Written once, read by everyone who needs to know.

Every tool that contributed to the project is recorded in one file. Reviewers, funders, auditors and automated checks all read the same record.

Claude CodeGitHub CopilotCursorWritten by handCI checksJournalsFundersAuditorsaidecl.yamlschema_versionprojectai_usage  tools  ai_proportiondeclaration

Declare, validate and keep it current.

Declare

State whether AI was used. If it was, list the tools and what each one was used for. Try it:

 

State the proportion

Record how much of the work was written by people, written with AI assistance or generated by AI, and how you estimated it.

21%with AI
  • 79% written by hand
  • 2% AI-assisted
  • 19% AI-generated

Figures from the pySDC project, estimated from git blame.

Validate

The validator catches mistakes that are easy to miss in review, such as a value the schema does not allow.

✕ aidecl.yaml
  Schema error at 'ai_usage.tools.0.hosting':
  'cloud' is not valid. Allowed values:
  'cloud_vendor', 'cloud_self_hosted',
  'on_premise', 'local_offline', 'unknown'

Keep it current

Update the declaration with each release and whenever a new tool is introduced. Set a review date, and the validator warns you once it has passed.

  1. v1.0declaration updated
  2. v1.1declaration updated
  3. v1.2declaration updated
  4. v2.0declaration updated

Run it in CI

Add two steps to a workflow. Validation errors appear as annotations on the pull request.

- run: pip install aidecl
- run: aidecl validate aidecl.yaml \
    --output-format github-actions

Stand behind it

Every declaration names the person or team responsible for it and the date it was made.

declared_by: Platform team

More projects are asking for disclosure.

A 2026 study examined how 1,000 popular GitHub repositories handle AI-assisted contributions.

0of the 1,000 repositories have a policy on AI contributions

0percent of those policies require contributors to disclose AI use

0percent of those policies require a human in the loop

One format for many kinds of work.

A code repository

The declaration is validated in CI alongside the tests.

A paper or report

The AI disclosure a journal asks for, kept alongside the manuscript.

A dataset

Records how much of the data is synthetic.

aidecl.yaml

A trained model

Names the base model and the tools used to build on it.

A research project

Shows which parts of the analysis were AI-assisted.

AI-generated

Images and media

States which images, audio or video were AI-generated.

How it supports EU AI Act documentation.

The EU AI Act regulates AI systems. A declaration file does not make a system compliant, and nothing here is legal advice. What the file does is keep information the Act asks for alongside the code.

Annex IV, point 2(a)
The technical documentation of a high-risk system must describe “recourse to pre-trained systems or tools provided by third parties and how those were used, integrated or modified by the provider”. The tools list in the declaration supplies that information.
Articles 11, 14, 47
If you classify a system as high_risk, the validator rejects the file until it links to the technical documentation and the declaration of conformity and describes the human oversight measures.
Article 50
Two fields record whether your product discloses AI-generated content and deepfakes to the people who see them. These obligations have applied since 2 August 2026.
Chapter V
For a system built on a general-purpose AI model, the file names the model's provider and links to the summary of its training data.

What the format does not do.

  • Declarations are self-reported. No tool yet measures whether the stated proportions are accurate.
  • Declarations are not signed. The schema has a field for a signature, but no tool creates or verifies one yet.
  • A declaration does not make a product compliant with the EU AI Act or any other regulation.

Planned, not yet available.

  • Estimating the proportions automatically from git history, Git AI notes and Agent Trace records.
  • A policy file that states what a project allows, so that a check can compare the declaration against it.
  • A GitHub Action that flags AI tools the declaration does not mention.

Add a declaration to your project.

$ pip install aidecl
$ aidecl init
$ aidecl validate aidecl.yaml
✓ aidecl.yaml

Get the command line toolRead the schemaCreate one in the browser