Does the EU AI Act apply to non-EU companies?
Yes, and here is the test
TL;DR
Whether you are in the US, Asia, Australia, the UK or the EU, the same rules reach you. What matters is where your output is used, not where you are.
If your AI content is produced in the US but your customer relies on it in Germany, you are in scope of the AI Act.
It is not where your company is.
It is not where your servers are.
Not where the model was trained.
It is where the content shows up.
But for words of reassurance, GDPR worked the same way in 2018 and we all survived that (somehow).
Article 50 of the AI Act has applied since the 2nd of August, 2026, so this is not something coming next year. The questions to ask yourself are the following.
The test
Do you have an EU entity, office or staff?
If the answer to any of them is yes, it applies to you. No argument, no grey area. You are established in the Union and that is the end of the question.
Do you sell to, market to, or publish content for people in the EU?
If yes, your output is being used inside the European Union, and you are in scope. EU pricing, a localised page, European ad targeting, an EU press list, EU customers. Any one of those is enough.
Do you just happen to have some readers in the EU, with no activity there?
If you can genuinely say they are not buying, quoting, or relying on your services or content, you fall into a grey area that most people oversell. It only changes when someone starts using or relying on what you publish, and at that point you are in.
If in doubt, just message us and we will clear up any obligations you may have, quickly.
But realistically, will the EU fine a random company in Denver?
Enforcement across borders is certainly slow, but what you can expect is the same pattern as GDPR. If you have an EU touchpoint, a seller or reseller, a subsidiary, a payment entity, that touchpoint has its own compliance to worry about, and that is exactly where an audit goes snooping around.
It probably will not be the regulator messaging you. It will be your customer, asking in a vendor or risk assessment how you handle the AI Act.
That question arrives with a deadline attached, usually a renewal date.
Where your exposure actually is
If you can, think through where your company may use AI. Exposure can come from something as small as a ChatGPT prompt writing a LinkedIn post, all the way to backend automations that sort CVs and emails, or a chatbot that never tells customers it is AI.
One warning on that list. Those are not all the same problem. Telling people your chatbot is AI, and labelling published text, are transparency duties under Article 50. Using AI to filter job applications is a high risk use with a far heavier set of obligations attached, and those have applied since the 2nd of August, 2026 as well. If you are doing the second one, do not file it next to the first.
Penalties for transparency breaches run to 15 million euro or 3% of worldwide turnover, and for SMEs and startups it is whichever of the two is lower. Getting there does not take a regulator taking an interest in you either. Anyone can file a complaint, including a customer, a competitor, or a reader.
What to do this week
- Write down which of the three questions you answered yes to, and the date you decided. A position taken in advance is a position. One taken after a complaint is an excuse.
- List every place AI touches something you publish or ship to Europe. Content, chatbots, translations, support macros, automations.
- For anything public facing, record the review as it happens. Who reviewed it, when, and what changed.
- Ask your agencies and freelancers one question in writing: did you use AI, and what did it do.
This was never a question about where your company is registered. It is a question about where your output lands.
Europe is slowing tech down in plenty of ways, but 450 million people is not a market you get to opt out of, and this is the cheapest part of it to get right.
Start with the part that takes ten minutes.
Art50 logs who reviewed what and when, seals it, and publishes the proof to a public transparency log that you do not control and neither do we. When the vendor questionnaire arrives, you send a file instead of writing an essay that you hope holds up.
Free for journalists, newsrooms, charities, NGOs and public bodies.
