AI Usage Policy

AIAdopt, version 2.2

In force: 3 October 2026 · Next review: Q1 2027

Why are we publishing this document?

AIAdopt sells EU AI Act compliance training to other organisations. It would be odd if we didn't have an AI Usage Policy ourselves. We have one, since 21 April 2026, and we publish it here because we think that is consistent: practising what we teach. This document is not a marketing piece, it is our working internal policy. Anyone who wants to see what workable AI governance looks like in practice can read along here.

1. Document control

2. Purpose

This policy describes how AIAdopt deploys artificial intelligence in its own operations, and which rules apply to everyone who uses AI tools on behalf of AIAdopt. It exists for two reasons: to ensure that AIAdopt itself practises what it teaches its clients, and to comply with the AI literacy obligation of Article 4 EU AI Act, as well as the voluntary governance commitments AIAdopt has entered into under the EU AI Pact pledges.

3. Scope and organisational context

AIAdopt uses AI systems intensively as a working tool. This policy describes the principles, responsibilities and controls for that use.

Persons covered by this policy

AI systems covered by this policy

4. Roles and responsibilities

Responsibility for AI governance lies with Rob Ummels. The governance roles are not divided among different people. The explicit rules and reviews in this policy support implementation and control.

5. Permitted use of AI

Permitted applications

Prohibited applications

6. Operational rules

The following rules apply to day-to-day AI use within AIAdopt. They arose during the build-up phase of the organisation and are formally recorded here.

Rule 1: Noticeboard test for input

Before entering information into an AI tool, the operator applies the noticeboard test: would I not mind if this information were posted on a public noticeboard? If the answer is no, the information must not be entered. This rule applies to client names, personal data, contracts, financial data, login credentials, internal strategy, and third-party information obtained in confidence.

Rule 2: Output review before publication or deployment

No AI output is published, deployed, sent to a client or otherwise used externally without review by the founder. The review checks factual accuracy, accuracy of quotations and source references, alignment with the house style, absence of hallucinations, and compliance with copyright and applicable law.

Rule 3: Transparency about AI use

Where AI has played a substantial role in externally published content, the AI use is disclosed. AI-generated images are given an indication of the tool used. Articles in the Insights section carry an attribution at the end of the article. Section 8 sets out when this disclosure is legally required and when AIAdopt applies it as a matter of its own policy.

Rule 4: No shadow AI

Only AI tools listed in the AI Tool Inventory may be used for AIAdopt work. Browser extensions with AI features, free trial accounts of new tools, and personal AI accounts not registered for business use may not be used for AIAdopt or client data. Adding a new tool requires an update to the inventory before first operational use.

Rule 5: Knowledge of prohibited practices

The founder keeps up-to-date knowledge of the practices prohibited under Article 5 EU AI Act, and ensures that no AI use within AIAdopt falls into a prohibited category. An annual review confirms this.

7. Data protection and confidentiality

All AI use must comply with the General Data Protection Regulation (GDPR) and with applicable Belgian data protection legislation. Specific rules:

8. Transparency about AI use (Article 50 EU AI Act and internal policy)

Article 50 EU AI Act sets out transparency obligations for certain AI systems and certain AI-generated or manipulated content, including direct interaction with AI, deepfakes and AI-generated or manipulated text published to inform the public on matters of public interest. For that last category, the disclosure obligation does not apply where the text has undergone substantive human review or editorial control and a natural or legal person holds editorial responsibility for the publication. AIAdopt additionally applies broader transparency on a voluntary basis, including where Article 50 does not require it.

This section describes how AIAdopt works. If our use of AI changes, this section changes with it.

Code of Practice. AIAdopt has signed Section 2 of the EU Code of Practice on Transparency of AI-Generated Content. Section 2 is aimed at deployers who use AI for content to which the disclosure obligations of Article 50, paragraphs 4 and 5, apply. This concerns the labelling of deepfakes and of AI-generated or manipulated text on matters of public interest. The European Commission publishes the list of signatories on its website. Signing is voluntary and is not a certification.

Editorial responsibility. Rob Ummels, founder of AIAdopt, bears editorial responsibility for the content AIAdopt publishes. He can be reached for this at info@aiadopt.eu.

Human review. Texts that are AI-generated or substantively edited with AI are published only after the founder has substantively reviewed the final text. That review is more than a sign-off. He checks the facts, the legal statements, the sources and the conclusions. Where necessary, he amends or rejects them. If AI still substantively changes the text after that review, the changed text is reviewed again before it is published. The check before publication also covers whether a separate AI label is needed and whether that label is displayed correctly. This review applies to the Dutch and the English versions. The French versions have been translated with AI without that substantive review.

Illustrations. The footer of aiadopt.eu states on every page that all illustrations on this site are AI-generated from human prompts, naming the tools used. Current tools: Google Gemini and OpenAI ChatGPT. AIAdopt applies this statement as standard, including where Article 50 does not require separate disclosure.

Labelling of images. For AI illustrations that, given their content, presentation and audience, do not constitute a deepfake, AIAdopt uses the central statement in the footer as voluntary transparency. In addition, under its own policy, AIAdopt labels every AI-generated or manipulated image that could be mistaken for real, even if it is not a deepfake in the legal sense. That label is clear and accessible and perceptible at the latest on first exposure. AIAdopt uses the EU icon or an equivalent AI label in line with the design and placement rules of Section 2 of the Code. The label is included in the content itself or shown through an equivalent solution. AIAdopt takes into account that the label should also remain visible when the content is downloaded or forwarded. The label has a text alternative for screen readers.

Labelling of audio and video. For audio that qualifies as a deepfake, AIAdopt uses an audible disclosure at the start. Where a screen is available, a visual label is added as well. For longer audio and video, the disclosure is repeated where appropriate and given again after an interruption.

Spoken course material. The trainings include a read-aloud function. The voice is AI-generated, in every available language. The button the learner uses to start the audio says so, before playback begins. Current tool: ElevenLabs. AIAdopt does not create a voice clone of any staff member, learner, client or other identifiable person, and does not present the voice used as that of an existing person.

Articles. Articles in the Insights section carry an attribution at the end naming the human author and the AI used. That attribution also states that Rob Ummels bears final editorial responsibility. AIAdopt discloses these AI contributions as a transparency practice, even where the text has been human-reviewed and the law does not require disclosure. French articles state at the top that they have been translated from Dutch with AI.

Reporting. Anyone who spots an error in AIAdopt content, a missing AI label or a label that is incorrect can report it to info@aiadopt.eu with the subject "AI content report". AIAdopt investigates every substantiated report without undue delay. If the report proves justified, the content is corrected or the label adjusted. The person who made the report receives feedback where usable contact details are available. AIAdopt cooperates with competent supervisory authorities in accordance with applicable law. Anyone who has reason to suspect an infringement of the EU AI Act can also lodge a complaint directly with a competent market surveillance authority, as provided for in Article 85 EU AI Act. A report to AIAdopt is not required for this.

Imitation of existing people. AIAdopt does not use AI-generated images or audio that imitate existing, identifiable people.

Interactive AI on the website. AIAdopt does not use a chatbot or interactive AI agent on the website. If one is introduced, it will be clearly labelled as AI before the interaction begins.

9. Human oversight (principle of Article 14 EU AI Act)

Although Article 14 EU AI Act legally applies to high-risk AI systems, AIAdopt voluntarily applies the underlying principle of meaningful human oversight to all of its AI use. Based on the current AI Tool Inventory, AIAdopt develops or uses no high-risk AI systems. The principle is applied as follows:

10. Incident handling

Incidents involving AI tools used by AIAdopt are handled in accordance with the AIAdopt AI Incident Procedure. Triggers in brief: any incident that may constitute a personal data breach under the GDPR, any AI incident that may fall under a legal reporting obligation (including, where relevant, Article 73 EU AI Act), any incident affecting the integrity or availability of the AIAdopt training platform, and any externally published AI output that later proves materially incorrect or harmful.

11. Training and competence

12. Review and update of this policy

13. References

Approved by:

Rob Ummels, founder, AIAdopt

rob@aiadopt.eu · https://aiadopt.eu

Place and date: Maaseik, Belgium · 3 October 2026