The EU AI Act for innovation teams: what changes when AI joins your workshops
The works council has heard that AI in a workshop tool might be a problem under the new EU rules and wants a statement before the next session. Most of what innovation teams do with AI in a workshop is low risk under the AI Act. A few duties do apply; this piece tells you which, in plain words.
What the AI Act changes for workshops
The AI Act (Regulation (EU) 2024/1689) entered into force in August 2024 and applies in stages. It sorts AI by risk: prohibited practices, high risk, limited risk with transparency duties, minimal risk. The prohibitions came first, the transparency and high-risk duties follow over the next few years; which of them already bind you today is a question for your legal team.
Three points matter for workshops. Emotion recognition in the workplace is a prohibited practice. AI systems used for decisions about employment, such as evaluating the performance or behaviour of employees, are high risk. And people must be told when they interact with an AI system, with AI-generated content identifiable as such (Article 50). The Act also asks for AI literacy among staff who operate AI systems (Article 4).
This is not legal advice, only a map of what to ask your legal team.
What this means for a workshop tool
Capturing and structuring what is said
Transcribing a session, tagging ideas, questions and decisions, clustering themes: not high risk. The tool works on content, not on people. The personal data side is covered in transcription and GDPR.
Scoring ideas
An AI that suggests scores for ideas against your criteria is not high risk in itself. The idea is the object, not the person. Keep it that way: the team can change the score, and people decide.
Writing summaries and playing a role
Summaries, action lists, an AI playing the sceptical customer to challenge an idea: again, not high risk. The output is a draft a human reads, corrects and signs off.
Where the line is
The line is crossed when the same data is used to evaluate people. Talk time as a performance figure. A score on the person who proposed the idea rather than on the idea. A report to a manager on who contributed least. Once an output feeds a decision about an employee, you are in high-risk territory. Emotion or mood detection from voice is out altogether.
Transparency and literacy
Participants should know an AI is working in the room, and be told before the session. AI output must be marked as AI output. And the team using the tool should know what it does and does not do; that is AI literacy in practice. Tools like IdeaHone mark AI content, treat AI scores as suggestions people can edit, and do no emotion detection.
The AI Act and GDPR overlap but are not the same. GDPR is about the personal data: whose voice, stored where, how long. The AI Act is about the system and how it is used. You need both answers.
A checklist for the works council and the DPO
- Write down what the AI does in your workshops. One page in plain words: it transcribes, clusters, suggests scores, drafts summaries.
- Confirm nothing evaluates individuals. No talk-time rankings, no per-person scores, no contribution reports.
- Confirm there is no emotion or mood analysis. Ask the vendor in writing.
- Tell participants and label the output. A short listening notice before the session, AI content visibly marked in the result.
- Keep a human decision for every score and summary. Someone with a name approves it.
- Brief the facilitators. One hour: what it does, where it fails, how to correct it. That hour is your AI literacy evidence.
- Keep the GDPR side in the same document. Consent for voice profiles, retention, deletion. One document, one conversation with the works council.
Ask the vendor for their AI Act statement. A serious one is one page long and says which risk class their use falls in and why. Twenty pages of marketing is not an answer.
Three mistakes in the compliance conversation
- Treating it as a legal problem only. The works council is asking what this thing does to their colleagues. Answer that first, then add the legal view.
- Letting a useful feature drift. A talk-time chart in the facilitator's view helps balance a session. Forwarded to a line manager, it becomes an evaluation of employees. Decide where that data stops.
- Promising more than the vendor confirms. Say "the vendor states", not "it never does".
AI that suggests, people who decide
IdeaHone turns what is said into cards, scores ideas against your criteria with written reasons and drafts the summary. Every suggestion can be edited; no decision is made by a machine.
Does the EU AI Act apply to AI workshop tools?
Yes. The AI literacy duty (Article 4) applies to organisations that use AI systems, and transparency duties (Article 50) apply to certain systems. Typical workshop uses such as transcription, clustering, summaries and idea scoring are not high risk in themselves. Check your setup with legal.
Is AI scoring of ideas high risk under the AI Act?
Not when the score is on the idea and people can edit it and decide. It is different when scores are attached to the people who proposed the ideas or used to assess their performance.
Do we have to tell participants an AI is used?
In practice, yes. People must be told when they interact with an AI system, and AI-generated content should be identifiable. Tell them before the session, mark AI output and check the details with your data protection officer.
