In this article8 sections
"We only use the free version" is not a compliance defence. That sentence gets said in boardrooms across the Netherlands and Belgium every week, and it is wrong. The free AI tools EU AI Act question has one clear answer: price is irrelevant. What matters is whether your staff are using AI systems in a professional context, and whether you have taken measures to support their AI literacy so they can use those systems responsibly.
What Article 4 Actually Says
Article 4 of the EU AI Act creates an obligation for both providers and deployers. As amended by Regulation (EU) 2026/1744, it says they "shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf", and that this "does not require providers or deployers to guarantee any specific level of AI literacy of any individual." So the duty is to take measures, fitted to your staff's knowledge and experience and to the context the AI systems are used in, not to reach an expert or certified level.
The obligation is not conditional on:
-
Whether you signed a contract with the AI provider
-
Whether you pay a subscription fee
-
Whether the tool is enterprise-licensed or consumer-tier
-
Whether you have an official procurement record for it
The AI Act defines a deployer in Article 3(4) as any natural or legal person that uses an AI system under its authority, except for personal, non-professional use. A company whose HR team pastes candidate CVs into ChatGPT's free tier is a deployer. Full stop.
Why "Free" Feels Like a Defence (But Isn't)
The intuition behind the "we just use free tools" argument is understandable. If there is no vendor invoice, there is no contract. If there is no contract, maybe there is no obligation. That logic works in some regulatory contexts. The AI Act is not one of them.
The Act is use-based, not procurement-based. Article 4 itself points to the context the AI systems are used in and the people they are used on, and Recital 20 describes AI literacy as the notions people need to make informed decisions about AI systems. None of those factors change based on whether your finance director paid for GPT-4o or is using the free version.
There is also a shadow IT dimension that makes free tools more risky from a compliance standpoint, not less. Paid enterprise tools typically come with data processing agreements, which you need anyway under the AVG/GDPR. Free consumer tools often do not. So the company using the free tier of a major AI model may simultaneously have an Article 4 problem and an Article 28 GDPR problem sitting right next to each other.
The Risk Profile of Common Free Tools
Let's be specific. These are tools that routinely appear in SME environments:
ChatGPT free (OpenAI): Consumer account, no DPA by default, conversations may be used for model training unless you opt out. Staff use it for drafting emails, summarising documents, generating policy text. From an AI Act perspective, any professional use makes you a deployer with an Article 4 obligation. If it touches HR decisions or credit assessments, the stakes are higher.
Gemini free (Google): Similar structure. Google's free tier for individual accounts is not covered by Google's enterprise data processing terms. If a manager uses it to evaluate performance notes or plan a redundancy consultation, that is a professional deployment.
Microsoft Copilot free (Bing Chat, now Copilot.com): Microsoft distinguishes sharply between its commercial Copilot (covered by enterprise terms) and the free consumer version. The free version does not fall under the Microsoft Product Terms DPA. Companies using it on work devices are in a gap.
In every case, the AI Act obligation exists independently of the GDPR gap. Even if you somehow resolved the data question, Article 4 still requires AI literacy measures for your staff.
What Documentation Does Article 4 Need?
Article 4 does not prescribe a format, and it does not require you to keep records. The Commission's AI literacy Q&A says no certificate is needed and that organisations can keep an internal record of trainings and other guiding initiatives. Records are simply how you show the measures you took. A sensible set:
-
A record of which AI systems staff are using, including free and informal tools, not just licensed software
-
Training or awareness sessions with attendance records
-
Internal guidance on acceptable use, covering what these systems can and cannot do
-
A named person or team responsible for AI oversight
For a company of 30 people, this does not need to be a 100-page policy. A two-page acceptable use document, a 45-minute staff briefing with a sign-off sheet, and a simple tool inventory are a reasonable start. But "we didn't know staff were using it" is a weak answer to a supervisory authority.
The timeline matters here. The AI literacy obligation under Article 4 has applied since 2 February 2025, and national market surveillance authorities supervise it from 2 August 2026. If your organisation has not yet documented its AI literacy measures, do it this month.
The "Shadow AI" Inventory Problem
Most SMEs have a formal software stack and an informal one. The informal stack is where free AI tools live. A staff member bookmarks Claude.ai at home and starts using it for work tasks. Someone in accounts discovers Perplexity. The marketing coordinator uses an AI image generator with a free account.
None of these show up in procurement. All of them create deployer obligations.
The practical starting point is not a policy document; it is an honest conversation with your team about what AI tools they are actually using. That conversation needs to happen before you can write the policy, because a policy that only covers tools you already knew about is incomplete.
A lightweight tool inventory does not require enterprise software. A shared spreadsheet with columns for tool name, use case, user group, and data types involved is enough to start. The point is visibility. Once you have visibility, you can assess which tools carry higher risk — for instance, any tool touching employee data, customer personal data, or consequential decisions.
Sector Specifics for Dutch and Belgian Deployers
In the Netherlands, the Autoriteit Persoonsgegevens is the intended market surveillance authority for most AI topics, with the AFM and DNB for financial firms, but it has no AI Act powers until the Dutch implementing law is in force. Under the draft, a shortfall in AI literacy leads to an order to fix it (a "last onder dwangsom"), not a fine. In Belgium, no market surveillance authority has been formally designated yet. Meanwhile, both countries' data protection authorities already supervise the GDPR side of AI use.
For HR Directors specifically: if your recruitment process involves any AI-assisted screening, shortlisting, or assessment, and you are using free-tier tools to do it, you have an Article 4 literacy obligation now, and you are in high-risk territory under Annex III, where Article 26 deployer obligations apply from 2 December 2027. Whether a general-purpose free tool used for screening counts as a high-risk system is not settled, so the safe course is not to use free consumer tools for these decisions at all.
One Action to Take This Week
Before you write a single policy document, do the tool inventory. Send a short internal survey — five questions, anonymous if needed — asking staff which AI tools they use for work tasks, how often, and for what purpose. Collate the responses. You now have the foundation for both your Article 4 documentation and your broader AI Act compliance roadmap.
The price tag on a tool has never determined your regulatory exposure. What determines it is how, and by whom, and for what purpose the tool is being used inside your organisation.
If you want a structured starting point, the free 2-minute compliance check at comply.khairos.ai maps your current AI use against your Article 4 obligations and flags where gaps exist — whether your tools are paid, free, or somewhere in between.
References
Part of our guide to AI Literacy & Article 4
Not sure where your company stands?
Answer 15 questions about how your team uses AI. You get a compliance score and the gaps to close first, as a PDF.