If your practice, engineering firm or construction company uses ChatGPT, assistants or AI automation, the EU AI Act already applies to you, though far less than most people think. Today it mainly requires three things: taking measures so the team knows how to use AI (what the law calls AI literacy), not using any of the prohibited practices and, since 2 August 2026, being transparent in specific cases, such as an assistant that chats with clients or a generated image that looks real. The most demanding obligations, for high-risk systems, have been postponed to 2 December 2027 and only cover specific uses, such as screening job applicants with AI.
This guide explains what role your company plays, the timeline, what applies now, what comes next, and a checklist to comply without fuss.
This guide is for information only and isn’t legal advice. The binding text is the one published in the Official Journal of the European Union.
What the Act is and what role your company plays
Regulation (EU) 2024/1689, known as the AI Act, regulates artificial intelligence systems according to the risk they pose. In July 2026 it was amended by Regulation (EU) 2026/1744, part of the simplification package known as the “Digital Omnibus”.
The Act mainly distinguishes two roles:
- Provider: whoever develops an AI system and places it on the market or puts it into service under their own name. OpenAI, Anthropic, Google and Microsoft are providers of their tools.
- Deployer: whoever uses an AI system in their professional activity. That’s the role of almost any practice or builder that uses ChatGPT, Copilot or an automation.
Purely personal, non-professional use is outside its scope. Use at work, including by sole traders, is inside.
One nuance: if your company develops a system, or commissions one and puts it into service under its own name, it may also have provider obligations. For systems that aren’t high-risk, these are mostly about transparency and training; even so, it’s worth clarifying with whoever builds the tool.
The timeline
| Date | What starts to apply |
|---|---|
| 1 August 2024 | The Act enters into force |
| 2 February 2025 | Prohibited practices and AI literacy (Article 4) |
| 2 August 2025 | Rules for general-purpose models, which their providers (OpenAI, Anthropic, Google…) must meet |
| 27 July 2026 | The Digital Omnibus amendment enters into force, rewording Article 4 and postponing high-risk rules |
| 2 August 2026 | Transparency obligations (Article 50) |
| 2 December 2027 | High-risk systems listed in Annex III, such as recruitment (postponed from August 2026) |
| 2 August 2028 | High-risk AI built into regulated products, such as machinery or lifts |
What applies now
1. Training your team in AI (Article 4)
Since February 2025, companies that use AI have had to make sure their staff know how to use it. After the 2026 amendment, Article 4 requires providers and deployers to take measures to support the development of AI literacy among their staff and anyone operating AI systems on their behalf, taking into account their knowledge, experience, education and the context of use. The text makes clear that it doesn’t require guaranteeing any specific level of AI literacy for any individual.
It’s an obligation to make an effort, not to achieve a result. For a practice or builder, in practice it means:
- A short training session on what AI does well, what it doesn’t, which data is never uploaded and how to check what it produces.
- A one-page usage policy. Our guide on ChatGPT and data protection suggests the six questions it should answer.
- A simple record: date, who attended and what was covered. It isn’t an official certificate (no mandatory certificate exists), but it shows the company has acted diligently.
The Commission and Member States must also support companies, with particular attention to SMEs, for example through practical examples and materials. And the training can often be funded: see how to fund AI training for your team.
2. Not using prohibited practices
Article 5 bans a small number of uses. Hardly any of them come up in the day-to-day work of a practice or builder, but one is worth knowing: recognising people’s emotions in the workplace is prohibited, except for medical or safety reasons. If someone offers you site cameras or software that claim to “read” workers’ moods, ask exactly what it’s for before you buy.
The other prohibitions (manipulation, social scoring, biometric categorisation to infer sensitive data, among others) are far from normal work in the sector.
3. Being transparent in specific cases (since 2 August 2026)
Article 50 has applied since 2 August 2026. For the sector, what matters is:
- Assistants that talk to people. If you launch an assistant that talks to clients or buyers (for example, for aftercare on a housing development), it must be clear from the first message that they’re talking to an AI, unless it’s obvious.
- Images, video or audio that look real. If you generate or alter content with AI that could pass as authentic, what the Act calls a deep fake, you must disclose that it’s artificial. For example, a “photo” of a finished home in a development that doesn’t exist yet, used in advertising. For clearly creative or fictional works, it’s enough to disclose it in a way that doesn’t get in the way of the presentation. A line such as “AI-generated image, for illustration” covers most cases.
The providers of the tools have their own obligations too, such as marking generated content so it can be detected. That isn’t something you have to do.
What’s coming: high-risk rules, postponed to December 2027
High-risk systems carry the most demanding obligations. The 2026 amendment postponed them to 2 December 2027 for the uses in Annex III, and to 2 August 2028 for AI built into regulated products.
For a practice, engineering firm or builder, the most likely high-risk cases are:
- Employment: using AI to screen CVs, select candidates, evaluate workers’ performance or allocate tasks based on their behaviour.
- Creditworthiness of individuals: for developers, using AI to assess buyers’ creditworthiness.
- Critical infrastructure: for engineering firms that design AI systems used as safety components in managing water, gas, heating or electricity supply, or road traffic.
If you use such a system from that date, you’ll have to use it according to the provider’s instructions, assign a competent person to oversee it, keep the logs it generates and, for employment, inform workers and their representatives before putting it into use.
Drafting minutes, specifications or estimates with AI is not high-risk.
Fines and supervision
Maximum fines are up to €35 million or 7% of worldwide turnover for prohibited practices, and up to €15 million or 3% for breaching other obligations, such as transparency. For SMEs, the lower of the two figures applies. Each Member State designates its own supervisory authorities.
For most practices and builders, the real risk isn’t a fine but using AI without judgement: uploading data that shouldn’t be uploaded, or signing documents nobody has checked.
The AI Act doesn’t replace the GDPR
They’re separate laws that apply at the same time. The AI Act regulates artificial intelligence systems; the GDPR regulates personal data, whether AI is involved or not. If a tool processes your clients’ data, you still need a data processing agreement, to know where the data is hosted, and to upload only what’s needed. We explain this in ChatGPT and data protection in a technical practice.
Checklist
- Make an inventory of the AI tools used in the company, including those people use on their own accounts.
- Note who uses them and for what.
- Write a one-page usage policy.
- Train the team and keep a simple record of the training.
- Check that no tool recognises workers’ emotions or falls into another prohibited practice.
- Label AI-generated images and videos that could look real.
- Make sure assistants that talk to clients say they’re an AI.
- If you use, or plan to use, AI in recruitment or to evaluate workers, prepare for December 2027.
Frequently asked questions
Does using ChatGPT to draft documents make my practice “high-risk”?
No. Drafting minutes, specifications, reports or emails isn’t a high-risk use. What does apply is the training obligation in Article 4.
Do I have to register my AI tools anywhere?
Not as a deployer of tools that aren’t high-risk. Registration in the EU database is for providers of high-risk systems and for public authorities that use them.
Do I have to tell my clients I use AI?
The AI Act doesn’t require it in general, only in the transparency cases above: assistants that talk to people and content that looks real. National rules can go further (Italy, for example, requires professionals to inform their clients). Either way, explaining to clients how you use AI and who checks what it produces tends to build trust.
Does it apply to me as a sole trader?
Yes, if you use AI in your professional work. The training obligation is met proportionately: in your case, training yourself and anyone who works with you on the tools is enough.
Do I need an official AI training certificate?
No. The Act doesn’t require any specific certificate. What matters is being able to show the company has taken reasonable measures.