The EU AI Act is here
New rules for AI in all 27 EU countries. Some already apply. The next wave lands on 2 August 2026, and it covers the AI tools your business uses every day.
Updated July 2026
Where do you stand?
What does your business do with AI?
What is the EU AI Act?
One law for AI across the whole EU, in force since August 2024 and switching on in stages until 2028. It treats AI the way Europe treats machinery or medicine. The more damage a use could do, the stricter the rules. The official text (opens in a new tab) runs long, but the parts that matter to most businesses fit on this page.
The law cares what AI is used for, not how clever it is. The same AI system is unregulated when it summarises meeting notes and heavily regulated when it screens job applications.
It splits the world into providers, who build and sell AI, and deployers, the businesses that use it. Providers carry most of the load. Where you're based doesn't matter. Sell into the EU, or have your AI's output used there, and the law applies.
The makers of the big models, like Anthropic (Claude) and OpenAI (ChatGPT), answer to a separate chapter that has applied since August 2025. That chapter is their job, not their customers'.
Does it apply to my business?
You use AI tools at work. You're a deployer, with the short list of duties further down this page. The main one is making sure your team understands the tools they use.
You sell an AI product. You're a provider. Chat assistants and drafting tools carry honesty rules from August 2026. AI that decides about people carries far heavier ones from December 2027.
You're UK-only. If you have no EU customers and nobody in the EU uses your AI's output, the Act doesn't reach you. The UK has no equivalent law either. The moment you sell into the EU, that changes.
Four levels of risk
Every use of AI lands on one of these. The higher the level, the heavier the rules.
Select a level
Transparency
AI that talks or createsChat assistants, writing tools, image generators. The rules here are about honesty. People must know they're dealing with AI, and AI-made content must be machine-readable as such.
Remi and most business AI tools sit hereWhat breaking it costs
Fines are tiered by what you broke, not who you are. National regulators pick the higher of a fixed sum or a share of worldwide turnover.
Using a banned practice
Breaking the product rules, including the honesty rules
Misleading a regulator
The small-business exception
For SMEs the cap flips, and you pay the lower (opens in a new tab) of the fixed sum or the percentage. As of July 2026 no fine has been issued anywhere in the EU. Nobody wants to be the test case.
What applies when
The EU amended the schedule in June 2026, pushing the high-risk dates back. These are the current dates.
Content-marking long-stop
Products already on the market get until this date to finish machine-readable content marking. Two new bans also begin, on AI built to generate child sexual abuse material or non-consensual intimate images.
What you need to do
If your business uses AI tools rather than builds them, the list is short.
Make sure your team understands the AI they use
In force since February 2025. No exam, no certificate. The people using AI at work know what it's good at and where it fails, and you could show how they know it.
Label AI content in two narrow cases
From August 2026, AI-written text published on matters of public interest with no human review must say so, and deepfakes must always be labelled. If a person reviews and edits before publishing, you're already covered.
Don't let AI decide about people
Screening CVs, scoring credit, setting someone's insurance price. From December 2027 these carry the strictest rules in the Act. Use AI to draft and research, and keep decisions about people with people, or take advice first.
And what you don't have to do
- No registering with a regulator
- No CE marking, audits or certificates
- No impact assessments
- No new privacy rules. GDPR already covers your data, and this law doesn't change it
Those duties belong to providers of high-risk AI. Everyday business tools aren't that.
Where Remi fits
Remi is an AI assistant for small teams. It reads documents, answers questions, drafts, and runs workflows. Under the Act that's a transparency-level product, built so that using it keeps you inside the rules by default.
You always know it's AI. Remi says what it is in the interface, and AI edits are labelled as AI edits.
A human approves actions. Before Remi acts outside your workspace, it shows you exactly what it's about to do and waits for a yes.
There's always a record. Every AI action is logged, so 'what did the AI do, and when' always has an answer.
Your data stays yours. Hosted in the EU, never used to train models, deleted when you leave.
Common questions
Read the law yourself
General information, not legal advice. Based on Regulation (EU) 2024/1689 as amended in June 2026.