Get involved

Anyone can suggest an incident, dispute an entry, or propose a change to the classification, and the reason for each change is published. Several of the ways to take part take 10 minutes or less.

This page is about the control gap record. ECL-1, the standard, has its own working group, open to anyone.

What the project needs now

Outside reviewers for the record’s version 1.0: No one outside G-14 has checked the entries yet. Before the record’s version 1.0, at least two reviewers outside G-14 will mark the gaps in each entry from its sources, without seeing the record’s answers or each other’s, and their agreement with each other and with the record will be published. Review entries for version 1.0
Incidents from more places: Most of the 32 incidents involve organizations in the United States. Incidents from other countries, sectors, and languages would show whether the same gaps occur there. Suggest an incident
Reviewers for the classification: The 13 gaps were worked out from the first 25 incidents and applied unchanged to every incident added since. People who investigate incidents, audit controls, or write specifications are well placed to say where a definition is too broad, too narrow, or missing. Review the classification

Short tasks

Suggest an incident

Send a description of what happened and the action that caused the harm, with at least one primary source, such as an official record, a court or regulator document, a company statement, or a first-hand account. The inclusion rules say what qualifies. Incidents that are considered and left out are listed on the record page with the reason.

Correct or dispute an entry

Each gap marked in an entry cites the passage in the sources that shows it. If a gap is wrong, a source has been misread, or a newer source changes the facts, send the entry number, the gap, and the passage where the source says otherwise. Disputes are settled against the sources, and the outcome is published with the next release.

Practice on an incident

The practice tool shows what happened in an incident from the record and asks which gaps the facts support. It then compares your answer with the record’s, gap by gap, with the evidence for each. It needs no account, and nothing you enter leaves your browser.

Longer contributions

Review entries for version 1.0

Reviewers go back to the primary sources and follow the published rules to decide which gaps each incident shows, without seeing the record’s answers. Reviewing one incident from its full sources takes between half an hour and an hour. The section on how the review works covers who can take part, how entries are assigned, and how reviewers are credited. To apply:

Read the classification and the rules for applying it.
Do at least five incidents in the practice tool, which reports how often your answers agree with the record.
Write to the editor with your agreement figure, the entries you practiced on, a line about your background, and any interests to declare.

Review the classification

Propose a new gap, a sharper definition, or a new rule for applying the gaps, with the incidents that support the proposal. Each proposal receives a written answer, published with the next version of the classification.

Map a specification

People who work on AARM, the Agent Control Standard, OpenShell, or another specification for runtime control can check how it is mapped to the gaps and send corrections, or propose a specification to add.

Take part in ECL-1

ECL-1 is a public standard, separate from the record, that G-14 maintains. The ECL-1 Working Group is open to anyone for free. Its procedures cover proposals, required reviews, written decisions, public sessions and release checks; later minor and major versions have public review windows. Use the working-group contact to take part, or the repository channels linked on that page when available.

Translate the 13 gaps

The gaps and their questions fit on one page. A translation is published with the translator’s name once a second fluent reader has checked it. The English text remains the reference version.

How to send a submission

Send each submission by email to ecl@g14.ai, with the subject line shown.

Submission	Subject line	Links
Suggest an incident	ECL incident	Write the email: Suggest an incident
Correct or dispute an entry	ECL record, with the entry number	Write the email: Correct or dispute an entry
Propose a change to the classification	ECL classification	Write the email: Propose a change to the classification
Correct a specification mapping	ECL mapping	Write the email: Correct a specification mapping
Offer a translation	ECL translation	Write the email: Offer a translation
Apply to review entries for version 1.0	ECL entry review	Write the email: Apply to review entries for version 1.0
Anything else	ECL	Write the email: Anything else

You can write in any language. The section on what happens to what you send explains how submissions are handled.

How the review works

Before version 1.0, at least two reviewers outside G-14 will mark the gaps in each entry from its sources, without seeing the record’s answers or each other’s. Their agreement with each other and with the record will be published, and each disagreement will be settled against the sources.

Who can take part: Anyone who can read an incident’s sources and apply written rules to them. Investigators, auditors, safety engineers, lawyers, journalists, and researchers are welcome, and no background in AI is needed. G-14’s employees and contractors cannot be reviewers.
Interests: Before starting, each reviewer declares their employer and any interest in the organizations and products named in the record, including work for a company involved in an incident, for one of its competitors, or for a vendor of controls of this kind. No reviewer is assigned an entry in which they have an interest. The interests of credited reviewers are published with their names.
How entries are assigned: The editor assigns five entries at a time, and reviewers do not choose them. Reviewers receive the sources and the classification, but not the gaps the record marked or another reviewer’s answers. Entries a reviewer has done in the practice tool are not assigned to them, because the tool has already shown them the record’s answers. Reviewers are asked not to open the record’s page for an assigned entry until they have finished it.
Time: About four hours for a batch of five entries, over two weeks. For each gap marked, the reviewer cites the passage in the sources that supports it.
Disagreements: When reviewers disagree with each other or with the record, the editor goes back to the sources to settle it and publishes the reasons, as for the close calls in this release. If the sources cannot settle it, the entry records both readings.
Credit: Reviewers who want credit are named in the release and in the data files, with the entries they reviewed.
When: The first round opens once enough people have applied to review each entry twice. The dates will be announced on this page and in the change log.

In your own work

Use the data: The record and the classification are published as JSON and CSV, with a schema and a static API, under CC BY 4.0. Cite the version you used. We will link to published work that uses the data if you tell us about it. Data and API
Teach with it: A one-hour session plan, a poster of the 13 gaps, and exercises built on incidents from the record, for a class, a team, or a workshop. All of it is free to adapt. Teaching kit
Check your own systems: The self-check takes one of your AI systems through the 13 questions, lists the incidents that had the same gaps, and produces a report for whoever runs the system. For buyers, the model clauses turn the same gaps into contract terms. Check a system
Updates and discussion: New entries are announced in the record’s Atom feed, and each release is listed in the change log. The editor also runs AI Governance & Occupational Risk | EHS/HSE Controls (LinkedIn group), where incidents, controls, and the classification are discussed.

What happens to what you send

The editor reads each submission. Changes to entries and to the classification go into the next release, and the reason for each is published in the change log or in the notes on the record page.
Your name is published only if you ask. A written answer may quote what you sent, without naming you unless you agree.
Your email address is used only to reply about what you sent. It is not added to any mailing list or used for G-14’s business.
Contributions are published under CC BY 4.0, like the site’s text and data.
Everyone taking part is expected to follow the code of conduct.

The governance page describes who maintains the project, how decisions are made, and how G-14’s interests are kept out of the record.