I only use ChatGPT: the duties you still have

01Deployer, not bystander

The AI Act distinguishes two main roles. Providers develop AI systems and put them on the market. Deployers use ready-made systems professionally under their own responsibility. As soon as you use ChatGPT for product copy, an AI tool runs in your customer service or an assistant sorts your mail, you are a deployer in the sense of the law. Purely private use does not count.

The good news: for deployers of ready-made everyday tools the duties are manageable and cheap to fulfil. You still should not ignore them. They are easy to implement and hard to excuse when none of it exists in a dispute.

02Article 4: the AI literacy duty

Since 2 February 2025 the AI Act requires that everyone working with AI in your business has sufficient AI literacy. The law prescribes neither a certificate nor a fixed number of hours. What matters is that your team understands what the tools in use can do, where they typically fail and which rules apply in-house. The bar depends on the use case: someone who only drafts texts needs less than someone using AI in sensitive processes.

• A short training session covers many cases: what AI can do, where its limits are, which data is off-limits.
• Document it: who was trained when, by whom and on which content.
• Repeat it when new tools or new people arrive.
• Since August 2026 national authorities are in charge of enforcement. Missing documentation is an unnecessarily bad start.

03Labelling in customer contact

If AI talks directly to your customers or creates deceptively real content, transparency duties additionally apply since 2 August 2026: chatbots must identify themselves as AI, and realistic AI images and videos need a label. I covered the details and the timeline in the wiki post on the AI Act transparency duties. For you as a user the core task is: go through every place where AI is visible to the outside, and add a clear notice there.

04Internal ground rules

Most problems come not from the law but from sprawl: everyone on the team uses some tool with some account, and nobody has the full picture. A simple internal rulebook on one page prevents exactly that.

• Which tools are allowed, and with which accounts.
• Which data may go in, and which never: customer data, credentials, trade secrets.
• Who reviews AI output before it gets published or sent to customers.
• Who the team turns to when in doubt.

05Responsibility does not move

Legally, an AI tool is a piece of equipment, not a scapegoat. If a generated product text states wrong measurements or the chatbot makes a promise that does not exist, it is not the tool provider who is liable to your customer, it is your business. That is why a human look belongs before every publication, especially for prices, product properties and legal statements.

The effort is smaller than it sounds: a tool list, a short dated training, one page of ground rules. That is one afternoon, and it puts you ahead of most.