AI Charter
Tools and providers
I use tools from Claude (Anthropic) and ChatGPT (OpenAI). I only adopt a new tool once I know where it stores the data I enter and what it uses it for.
Purpose and scope
I use AI for writing code, debugging, documentation and drafts, so I can work faster and better. That means fewer billed hours for the same result. Decisions and responsibility do not belong to the AI.
Data
I may load a client's code into an AI tool, but never personal data, passwords, access keys or live databases. Where needed, I redact.
Experts
I do not give legal advice. If a task raises legal, data protection or copyright questions (for example under the GDPR or the EU AI Act), I flag it and recommend involving an expert. The decision is the client's.
Review
I review, understand and test every piece of AI-generated code before handing it over. My name is on it; I am responsible for the result.
Prompting rules
I always give the AI the goal, the context the task needs and the boundaries, and I check the result. I do not pass on anything I do not understand.
Decisions about people
I do not leave decisions about people (such as hiring, credit or exclusion) to AI, and I do not build such systems without human review.
Chatbots and agents
Any chatbot or agent I build tells users they are talking to AI. It only accesses the data and systems its role requires, and anything involving money, contracts or personal data needs human approval.
Transparency
I use AI tools in development, for example to draft code snippets and outlines. I review and test all delivered work myself, and I am responsible for the result. I do not load client data or access credentials into AI tools.
Prohibited uses
I do not hand over AI-written code that I have not reviewed, understood and tested. I do not use AI to create fake reviews, references, non-existent people or misleading content.
Ethical limits
I do not build anything that collects or analyses personal data with AI without users knowing, even where it would be technically and legally possible.
Incidents
If an error turns up in AI-assisted work, I tell you right away, and if it is my fault, I fix it free of charge. If it affects personal data, I help with the notification.
If I build AI systems
I also build AI-powered systems for clients. In that case the client is the deployer: at handover I document which model and provider the system uses, what it does and does not do, what I know and do not know about how it works, and how to monitor it.