EU AI Act August 2, 2026: what changes for users and businesses?
From August, the European Union switches on the full sanctions regime for providers of large AI models, while users get clearer rules on transparency and copyright.
Does the EU AI Act actually affect you if you use ChatGPT from outside the bloc? Directly, no. Indirectly, yes, because major model providers change behaviour globally to avoid maintaining separate regimes. From August 2, 2026 the European Commission gets full supervisory powers, and fines for non-compliance can reach three percent of a company’s annual global turnover.
The law has been partially in force for some time already: obligations for providers of large language models took effect on August 2 of last year. The August 2026 date is the point when the Commission can actually impose sanctions, which is the line companies are watching.
What “general-purpose AI models” means
By Brussels’ definition, these are models that can perform a wide range of tasks. ChatGPT, Claude, Gemini, Llama all qualify as general-purpose. The rules for them are stricter than those for small specialised models.
There is also a “systemic risk” subcategory for models above a certain compute threshold. Today, that only covers the largest models from OpenAI, Anthropic, Google and Meta. They have to perform additional risk analysis, report serious incidents on a regular cadence, and put formal safety measures in place.
What a company has to do
If your company uses an off-the-shelf AI model (ChatGPT, Claude API, Gemini), you have no direct obligation regarding the model itself. But if you sell an AI-based service to end users in the EU, you must transparently disclose that the content is AI-generated, especially for images, audio and video.
If your company trains its own model, the list of obligations is longer:
- Documentation of the training data, respecting EU copyright law.
- A summary of the content used for training, made publicly available.
- Technical documentation about the model on request from the EU AI Office.
- A policy aligned with the Copyright Directive.
Models that were on the market before August 2, 2025 got a transitional period running to August 2, 2027.
The May 2026 simplifications
The Council and Parliament of the EU reached an agreement on May 7 to simplify some rules. Supervision of systems based on general-purpose models has been centralised at the EU AI Office, reducing fragmentation between national regulators. That is good news for companies that had been bracing for twenty-seven different interpretations of the same rules.
What it changes for the average user
Even where the AI Act does not apply directly, when Anthropic or Google changes default model behaviour to satisfy the EU, those changes show up everywhere, because companies rarely maintain parallel versions for different markets.
You can see this in two concrete places:
- Watermarks on AI-generated images, which Google and OpenAI have started embedding automatically.
- Mandatory transparency when a chatbot is talking to a user, which now shows up on all major platforms in the form of a “You are speaking with an AI” notice.
A second direction of impact: companies working with EU clients will have to adapt their own AI systems to the same standards, because an EU client will not sign a contract with a provider that does not comply with the AI Act.
What is next
The Commission has already published guidance for providers of general-purpose models, a practical walkthrough of the obligations. The full regime kicks in on August 2, and one more technical clarification from the EU AI Office is expected on the penalty mechanism. The date companies are already pencilling in is August 1, the last business day before the full regime starts.
Practical step for a small business
If you run a small business that uses an off-the-shelf AI tool for content, marketing or customer support, do three things before August 2:
- Check whether your tool (ChatGPT, Claude, Gemini, Copy.ai) has a clear statement of AI Act compliance. Major names already do.
- Add a clear notice on your site when a user is talking to a chatbot or seeing AI-generated content.
- If you are routing EU citizens’ personal data through an AI system, check whether your tool respects GDPR. The AI Act builds on GDPR, it does not replace it.
For most small businesses, this is not a heavy lift. The large players have already adapted their tools, your job is just to use them transparently.
