AI Usage Guideline#

Usage of Generative AI
status: valid
tags: process_management
version: 1
contained by: wf__cr_mt_process_mgt_strategy

This guideline defines how the Eclipse Generative AI Usage Guidelines, made applicable by Policies, are applied in the project. It applies to every contribution to the project, independent of the repository and the kind of content, including issues, pull requests and review comments.

The Eclipse guidelines remain the governing document. This guideline only makes them concrete where the Eclipse guidelines leave a choice to the project.
gd_guidl__ai_usage
Process Guideline

Responsibility#

  1. The contributor (Contributor) shall be responsible for every change they submit, independent of whether it was created with the help of a generative AI tool. The contributor shall have reviewed and understood every change and shall be able to defend it in the review.

  2. The use of a generative AI tool shall not reduce the scope of the review, the acceptance criteria or the automated checks that apply to a contribution.

Disclosure#

  1. The contributor shall disclose the use of a generative AI tool, if the AI contributed in a way for which a human would be given credit.

    No disclosure is required for trivial assistance, for example spelling and grammar corrections, formatting, code completion of single statements or rote boilerplate that the contributor modifies and supplements.

  2. The disclosure shall be given where the AI-assisted content is created:

    • for changes in a repository, as Assisted-by: trailer in the message of every affected commit;

    • for issues, pull request descriptions and comments whose text is AI-assisted, as Assisted-by: line in that text.

  3. The AI tool shall be identified in the following format, as complete as possible:

    Assisted-by: <Provider> <Model> (<Version/ID>)

    Examples:

    Assisted-by: GitHub Copilot (GPT-4o-2024-08-06)
Assisted-by: Anthropic Claude Opus 5.5 (claude-opus-5-5)

    If several AI tools were used, one Assisted-by: line per tool shall be given.

  4. A file that is largely or entirely AI-generated shall carry, in addition to the standard copyright header of the project, an AI disclosure block directly below it:

    AI Disclosure: This file was largely AI-generated. The AI-generated
portions are made available under CC0-1.0 and not subject to the
project's license. The human contributor has reviewed and verified
that the content is correct.
SPDX-License-Identifier: Apache-2.0 AND CC0-1.0
Assisted-by: <Provider> <Model> (<Version/ID>)

    The copyright header of the project shall stay unchanged, so that the automated copyright checks keep working.

Review and Merge#

  1. Every merge into a repository of the project shall be approved by a human committer (Committer). AI agents or bots shall not push directly to protected branches or merge pull requests.

  2. A committer may reject an AI-assisted contribution that they do not understand or are not willing to maintain, or request that it is split into smaller pull requests.

  3. A committer shall request the missing disclosure, if a contribution appears to be AI-assisted and is not disclosed.

Functional Safety and Security#

  1. AI-assisted content in safety or security related work products shall follow the same process, review and verification requirements as any other content. No additional rules apply.

  2. AI tools used to create or verify work products shall be identified and evaluated like any other tool according to Tool Management (Evaluate Tool and Update To...). A qualification of AI tools is not intended. Therefore the output of an AI tool shall not be used as work product or verification evidence without human review.