The EU AI Act for B2B teams that automate with AI
03 November 2026
Most B2B automation is nowhere near the strict end of the EU AI Act. A few uses are, and they tend to appear when a team starts automating HR or finance. This guide separates the two, with the dates and article numbers checked against the text of the regulation. It is not legal advice. Your counsel decides what applies to your company.
What is the EU AI Act?
Regulation (EU) 2024/1689 is the EU law that regulates AI systems by risk. It bans a short list of practices, sets strict rules for high-risk uses such as recruitment and credit scoring, and adds transparency duties for chatbots and generated content. Most business automation falls into the lower tiers.
In practice there are four tiers:
- Prohibited practices (art. 5). Social scoring, manipulative techniques that distort behaviour, and emotion recognition in the workplace or in education, among others. These are banned outright.
- High-risk systems (art. 6 and annex III). Uses in areas such as employment, credit, education and essential services. These carry heavy obligations.
- Transparency duties (art. 50). People must be told when they interact with an AI system, and generated content must be marked.
- Everything else. No specific obligations beyond the general ones, such as AI literacy in art. 4.
Fines follow the tiers. Breaking a prohibition can cost up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher. Most other operator obligations are capped at EUR 15 million or 3%. For SMEs and start-ups the lower of the two figures applies (art. 99).
When do the rules apply?
The dates changed this summer, so older articles are wrong on high-risk. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published on 24 July 2026 and moved the high-risk dates.
| Date | What applies |
|---|---|
| 1 August 2024 | The AI Act entered into force. |
| 2 February 2025 | Prohibitions (art. 5) and AI literacy (art. 4). |
| 2 August 2025 | Rules for general-purpose AI models, governance and penalties. |
| 2 August 2026 | General date of application, including the transparency duties of art. 50. |
| 2 December 2026 | Generative systems already on the market before August 2026 must comply with the marking duty in art. 50(2). |
| 2 December 2027 | High-risk systems listed in annex III: recruitment, credit scoring and the other areas below. |
| 2 August 2028 | High-risk systems that are safety components of regulated products (annex I). |
The original text set the annex III date at 2 August 2026 and the annex I date at 2 August 2027. If you read that somewhere, it is out of date.
Does the EU AI Act apply to my company?
Yes, if you build AI systems for the EU market or use them in your business in the EU. The law calls the first group providers and the second deployers. Email classification, outreach and document assistants usually carry light duties, mainly telling people when they deal with an AI.
Two roles matter. A provider develops an AI system, or has one developed, and puts it on the market under its own name. A deployer uses an AI system in its own business. An agency that builds a chatbot for a client is a provider. The client that runs it is a deployer. A team that builds its own assistant and uses it internally is both.
Where typical automation falls
This is our reading of common B2B automations, not a ruling, and it depends on how the system is used.
Email classification and drafting. Sorting inbound email and drafting replies for a person to approve is generally minimal or limited risk. If customers chat with an AI directly, art. 50(1) applies.
Outreach and lead qualification. Writing and sending B2B messages is generally limited risk. A conversational agent that talks to prospects must make clear it is an AI unless that is obvious from the context (art. 50(1)). Outreach also has to meet GDPR and national marketing rules, which are separate.
RAG assistants. An assistant that answers from your documents is limited risk. The provider must design it so people know they are talking to an AI. A RAG chatbot that hands over to a person when it is unsure is also easier to defend on accuracy.
Content generation. Providers of systems that generate text, images, audio or video must mark the output as artificially generated in a machine-readable format (art. 50(2)). The duty on you as a deployer is narrower: deep fakes must be disclosed, and so must AI-generated text published to inform the public on matters of public interest, unless a person has reviewed it and holds editorial responsibility (art. 50(4)). A marketing post that a person edits and approves generally falls outside the second rule.
In every case the information has to reach people clearly, at the latest at the first interaction (art. 50(5)).
What is high-risk and often catches SMEs
Annex III lists the high-risk uses. Three catch ordinary companies:
- Recruitment and selection (annex III, point 4a). Systems that place targeted job ads, filter applications or evaluate candidates.
- Managing employees (point 4b). Systems that decide on promotion, termination or the terms of work, allocate tasks based on behaviour or personal traits, or monitor and evaluate performance.
- Credit scoring (point 5b). Systems that evaluate the creditworthiness of individuals, except those used to detect financial fraud.
A CV-screening workflow built in an afternoon is generally a high-risk system. Art. 6(3) allows an exception where the system only does a narrow procedural task or a preparatory step and does not materially influence the decision. That exception never applies if the system profiles people. Do not rely on it without checking.
The deployer duties for high-risk systems include human oversight by trained staff, monitoring, keeping the system’s logs for at least six months, and telling workers before a high-risk system is used at work (art. 26). The provider duties are heavier still.
Art. 5 also bans emotion recognition in the workplace, except for medical or safety reasons, so a tool that scores the mood of staff on calls is out.
What is the difference between GDPR and the AI Act?
GDPR protects personal data and applies whenever you process it. The AI Act regulates the AI system itself, by risk level, whether or not personal data is involved. An automation that reads customer email has to meet both.
The GDPR checklist for AI automation covers data location, processors and retention.
What do we have to document?
For most automation: a list of your AI systems with their purpose and risk level, a note on where people are told they deal with an AI, and a record of staff training. High-risk uses need much more, including logs kept for at least six months and human oversight.
In practice, three things:
- A register of your AI systems. What each one does, who uses it, what data it sees and which tier you place it in, with the reason.
- A record of where people are told. The chatbot greeting, the label on generated content, the line in the email footer.
- A training record. Art. 4 requires providers and deployers to take measures to support AI literacy of the staff who work with AI systems. It does not require a specific level for each person.
For a system in the high-risk tier, add the oversight owner, the log retention period and the instructions for use that the provider supplies.
What a team can do this week
- List every AI tool and automation in use, including the ones individuals signed up for themselves.
- Mark any that touch recruitment, employee evaluation or credit decisions. Those need a closer look first.
- Check that each chatbot and agent tells people they are dealing with an AI.
- Decide who reviews AI-generated content before it is published.
- Write down the training your team has had, and fill the gaps.
If you do not know what is running in your operations yet, the free automation audit starts by mapping it. Our AI consulting work covers the next step: deciding what to build and how to document it.