Skip to content

Question

Do I need an AI policy for my business?

No, a written AI policy is not required anywhere in the EU, but AI literacy is: if your staff use ChatGPT, Copilot or another AI tool for work, you make sure they understand what it does.

What the EU AI Act asks per way of using AI and since when, what it costs a business of ten, and a policy on one page.

Do I need an AI policy: an empty meeting room with a glass wall and a long table, where you agree how your business uses AI

By , Founder and CEO

3 min read

No policy required, but these duties

  • Nobody in your business uses AI toolsNoThen there is nothing to arrange. Do check: many people use ChatGPT on their own.
  • Your staff use ChatGPT, Copilot or GeminiNoRequired since 2 February 2025: AI literacy, with explanation and agreements. No fixed format, no certificate.
  • A chatbot on your site talks to customersNoRequired since 2 August 2026: customers know they are talking to AI, unless that is obvious.
  • You publish images or video that look realNoRequired: say they were generated or edited with AI.
  • AI helps assess applicants or staffNoFrom 2 December 2027 the high-risk rules apply: human oversight, logs, informing your people.
  • You build AI software for othersNoThe duties of a provider apply, heavier than a user's. Get legal advice.
On this page
  1. 01What AI literacy asks in practice
  2. 02What it costs a business of ten
  3. 03Why a policy is still smart

What AI literacy asks in practice

Since 2 February 2025, article 4 of the EU AI Act asks businesses that use AI systems to take measures so the people working with them know enough about what those systems do. In July 2026 the act was amended through the Digital Omnibus, Regulation (EU) 2026/1744: the duty was simplified and the Commission and member states took a bigger role in promoting AI literacy, but it stayed. The Commission's AI literacy Q&A of October 2026 says there is no fixed level, no certificate and no written policy required. If you want one anyway, start from the AI policy template. What counts is that you can show you did something: explanation, agreements, a short training. What the AI literacy requirements ask after the omnibus, and how to document them, is in AI literacy requirements.

The act applies across the EU, and to businesses outside it whose AI output is used in the EU. National supervisors may act since 2 August 2026, though not every country had named one by then. The act ties no fine of its own to AI literacy; for prohibited practices penalties reach 35 million euro or 7 percent of turnover, for other duties 15 million or 3 percent, and for a small business the lower of the two applies. The real risk for a small business lies more in a client file pasted into a free chatbot than in a fine. The EU AI Act checker shows in a few questions which rules apply to you.

What applies when

  1. 2 Feb 2025Literacy and bansExplanation for AI users; manipulation banned.
  2. 2 Aug 2026TransparencyMake chatbots and deepfakes recognisable.
  3. 2 Dec 2026MarkingOlder AI tools mark their output too.
  4. 2 Dec 2027High riskStaff, credit, education.
  5. 2 Aug 2028ProductsAI in regulated products.
The dates of the EU AI Act after the July 2026 omnibus. Most small businesses use AI as a tool and sit at literacy and transparency; high risk only arrives at the end of 2027.

What it costs a business of ten

An accounting firm of ten lets everyone work with ChatGPT and wants to do it properly:

ItemSumCost
ChatGPT Business, so input does not train10 seats times 21 euro times 12 months2,520 euro a year
Explanation per employee10 people times 2 hours times 50 euro1,000 euro, once
Writing a one-page policy4 hours times 50 euro200 euro, once
Yearly refresher10 people times 1 hour times 50 euro500 euro a year
First year together4,220 euro

Most of it is the licence, and you pay that not for the law but for your data: on business plans what you type does not train by default, on free and personal plans it does until you switch it off. Plan prices are in ChatGPT pricing; whether ChatGPT is safe enough for client files is on is ChatGPT safe for business; comparing tools, see ChatGPT vs Claude; to keep data in Europe, which chatbot is European.

Why a policy is still smart

A page of agreements is the simplest way to show you worked on AI literacy. It also protects you against what goes wrong more often than a fine: an employee pasting a client file into a free chatbot, or a quote with an invented figure going out. If AI changes how your team works, check your local rules on informing and consulting staff, and whether your privacy policy needs an update. How to build the policy is in the EU AI Act for business guide; chatbot details in EU AI Act transparency obligations, and why high risk moved in EU AI Act high-risk delay.

An AI policy on one page

  • Which AI tools are allowed, with which account: the company account, never a free private one for work.
  • Which data never goes in: personal data of clients and staff, contracts, figures that are not public.
  • Who checks what AI produces before it reaches a client.
  • When you say something was made with AI: in a chatbot always, for images that look real too.
  • A short guide per tool for newcomers, and who answers questions.

AI for business

How VBS brings in AI with agreements

We set up AI tools that fit your work, with the agreements and the human check built into the tool rather than on paper: which data may go in, who reviews, what the client sees. That is AI for business; automating recurring work is AI automation. We built AMOSI, our marketing software that went live in July 2026, the same way: nothing went out before a person had seen and approved it.

AI in your business, with agreements that work?

We set up AI tools that fit your work, with the agreements and the human check built in rather than on paper.

Optional, if you prefer a call.

What do you want to achieve, and what is already there?

We only use your details to reply to you.