AI Transparency Statement
Jockkie Studio · Jock Oosterveer · Last updated: 17 October 2026
This page explains how I use AI: in the images, clips and films I make, and in the AI I set up for teams. It says which EU rules apply to me now, what I already do before stricter rules start in 2027, and how you can check it. Everything I make for clients, I make as Jockkie Studio. Until October 2026, Jockkie Studio was called AIPION. This page replaces the AIPION version of 6 August 2026.
In short
• AI is my instrument, not the author. I direct, check and approve every piece.
• Every piece is checked against your own source before it reaches you.
• AI work is marked as AI work. Every photorealistic person is labelled "Made with AI".
• For every project I keep a log of which AI model, and which version of it, made each piece. You can ask for it at any time.
• Your material only goes where you allow it, and you choose how long I keep it.
• In the AI assistants I set up, nothing goes out until a person on your team approves it.
1. How I work with AI
AI is an instrument, never the author. Ideas, judgement and final decisions stay with a person. Nothing made with AI goes out as finished work without my review.
I don't only make AI images and films. I also set up AI that takes work off a team's hands. In both, I start from the person and their work, not from the tool: what the person keeps doing, where they decide, and whether they still understand the work. AI prepares and checks; people decide. The question I ask at every step: does this make the person stronger, or unnecessary?
A person can only really approve AI work if they have four things: the information to see why the AI proposes something, the time to disagree, the authority to say no, and the practice to do the work without AI. I keep those four in mind when I design an approval step.
2. What I do in practice
• Your source first. Every claim in your film or images comes from a line in your own source, such as a spec sheet. Where your sources disagree, or a claim carries risk, you confirm it in writing before it is used.
• Labels. Images, video and audio made with AI are marked as AI work where I publish them. Every piece in my portfolio says which tools made it. Every photorealistic person is labelled "Made with AI", so a generated face is never passed off as a real person. I don't show children in client work.
• A notice that it was made with AI, where needed. Before delivery I check whether the work shows a real or generated person, the voice of one of your staff, or a realistic version of an existing product. If so, I decide whether a notice is needed and where it goes: in the image, in the voice-over, or where the work is published. If a notice is needed, the work is not delivered without it. This check is broader than the law, which asks for a notice on deepfakes (see section 4).
• A log per project. For every AI generation I log the model, its version, the account it ran in and the date. Failed attempts are logged too. You can ask for the log at any time.
• Voices. The voice of one of your staff is only used with that person's written consent.
3. Your material
• You choose what AI may do with your material, in the intake form. Four levels: open (AI services in the cloud may be used), under NDA (images and video only in FLORA if the NDA allows it), local (AI only on my own computer), or no AI at all.
• In the cloud, images and video with your material are made only in my paid FLORA workspace. According to FLORA's own terms, nothing you upload or generate there is used to train AI models, also not by the AI companies FLORA works with. I never put your material directly into the website of an AI model.
• I keep your material for 12 months after delivery, and then remove it. You can choose a shorter period, including "delete right after delivery". The planned removal date goes into the control report of your project.
• The services that may handle your material, AI services and others, are listed in my supplier file, available on request.
I'm a FLORA Creative Partner and a Monid affiliate: I earn when people sign up through my links.
4. The EU rules that apply to me now
The EU AI Act gives people roles. I'm a deployer: someone who uses AI in their work. I'm not a provider: I don't make AI models or AI tools and sell them under my name. The assistants I set up run on existing software and existing language models.
What is required, and of whom
• Measures so that the people who work with AI understand it: Required since 2 February 2025. I have no staff, so I apply it to myself.
• A notice on deepfakes, and on AI-made text about public affairs that no person checked: Required of me, since 2 August 2026.
• Telling people they are talking to an AI system: Required of the companies that make AI tools.
• A hidden digital mark in what AI makes, so software can recognise it as AI-made: Required of the companies that make AI tools. Tools that were already on the market before 2 August 2026 have until 2 December 2026.
• Logs, oversight by trained people, documentation and handling incidents, for high-risk AI: Required for high-risk AI from 2 December 2027. Not my work. I do my own version voluntarily, see section 5.
• A label on all my AI work, and "Made with AI" on every photorealistic person: Voluntary. The law asks less.
• The production log, the control report, the four levels for your material: Voluntary.
• Keeping personal data no longer than needed: Required by the EU privacy law (GDPR). For other material, voluntary.
AI knowledge. A business that uses AI must take measures to support the AI knowledge of the people who work with AI on its behalf. The law does not set a fixed level. I have no staff, so I apply it to myself. What I do: I keep a register of the AI models I may use for client work, per level of confidentiality, and a new model only goes on it after I approve it. I test image and video models in my own work and write guides about them on my blog.
Notices. As a deployer I must give a clear notice:
• on a deepfake: AI-made image, sound or video that looks like real people, things, places or events and could be taken for real. The notice must be there the first time someone sees or hears the work. For clearly artistic or fictional work, a lighter notice is enough, as long as it is there;
• on AI-made text that is published to inform the public about public affairs, unless a person checked it and takes responsibility for it.
My check before every delivery, in section 2, is how I meet this.
Banned uses. My work is images, clips and films, and assistants for a team's own work processes. None of it does anything the AI Act bans, such as manipulating people in ways that harm them, exploiting people's vulnerabilities, social scoring, or reading emotions at work or in school. From 2 December 2026 the ban also covers AI that makes sexual deepfakes of real people without their consent, and AI that makes child sexual abuse material.
Personal data. The GDPR applies whenever personal data is involved, for example images or voices of your staff.
5. Ahead of 2027: what I already do voluntarily
On 27 July 2026 a change to the AI Act took effect. It moved the strictest rules to later. The rules for high-risk AI now start on 2 December 2027. High-risk AI is, for example, AI that selects job candidates, assesses staff, or decides who gets an essential service. For AI built into products such as machines and medical devices, the date is 2 August 2028. These rules ask for four things: a log of what the AI did, oversight by trained people, documentation, and the handling of incidents.
My work is not high-risk, and I don't build or set up high-risk AI. So these rules don't apply to me. I already work along the same lines, because a buyer, a lawyer or a privacy officer asks these questions anyway. This is voluntary. Here is what I do now, per point.
1. A log of what the AI did.
• Every project has a production log: each AI generation with its model, version, account and date, including failed attempts.
• The log stays in the project archive after the project closes.
• For an assistant, the one-page agreement names who reads back its work.
2. Oversight by people who can step in.
• Nothing reaches you without my approval.
• In an assistant, a person on your team approves. The assistant doesn't get the tools to send anything on its own.
• The people who approve learn how to check its work, correct it, answer its approval requests and pause it.
3. Documentation of each assistant.
• Every assistant comes with a one-page agreement: the work it does, what it may never do, which language model it uses and why, who may use it, who approves and reads back, how to pause it, and who manages it.
4. Watching for mistakes, and acting on them.
• Every mistake in my work goes into an error log, with the date, the step and the cause.
• A mistake that happens twice becomes a fixed rule in my checks.
• An assistant can be paused with one command. In its trial week, every difference between its work and your team's own work goes on an error list.
6. Your rights and questions
• You can ask which tools, and how much human work, went into any piece I made.
• You can ask for the production log of your project.
• You can ask me to look again at any AI-made content about you or your work.
• Questions or complaints: info@jockkie.com. About personal data, you can also contact the Dutch Data Protection Authority, the Autoriteit Persoonsgegevens (autoriteitpersoonsgegevens.nl).
7. Words used here
• Deployer: someone who uses an AI system in their work. That is me.
• Provider: a company that makes an AI system or model and puts it on the market under its own name.
• Deepfake: AI-made image, sound or video that looks like a real person, thing, place or event and could be taken for real.
• High-risk AI: AI used for important decisions about people, such as hiring, or built into products where safety is at stake, such as machines.
• Production log: the log of a project, with every AI generation.
• Control report: my one-page report per project on independence, business, and data and AI.
8. Changes to this page
I update this page when the rules change, when my way of working changes, or when someone points out a gap. Every material change gets a date and a short note here.
• 17 October 2026: Rewritten for Jockkie Studio. Replaces the AIPION version of 6 August 2026. Added the change to the AI Act of July 2026, the new dates, and the section "Ahead of 2027".
9. Sources
The official texts behind this page: