EU AI Act: definition, duties and impact on assistants
Definition
The EU AI Act is the European Union regulation that governs the use of AI systems by risk. The higher the risk, the stricter the duties, from banned practices through high-risk obligations to transparency rules for general systems. It applies to providers and deployers, so it also binds companies that run an AI assistant or a company brain purely for internal use.
Why the Term Matters
Many teams roll out an assistant without knowing which class it falls into. Yet that decides whether a short label is enough or whether documentation, oversight and risk management are needed. Who thinks about the EU AI Act early builds permissions, logs and human approvals in from the start, instead of retrofitting them expensively later.
When It Is Relevant
- An assistant or agent goes into productive use in the company.
- The system touches hiring, creditworthiness or similarly sensitive fields.
- Users should know they are talking to an AI and not a human.
- You buy in an AI product and need clarity on provider duties.
When Not
For a pure experiment in a closed circle with no real decisions, the heavier duties usually do not apply. And the AI Act replaces neither data protection law nor employment law nor sector oversight, it comes on top and does not solve any single compliance problem on its own.
Industries That Use It
- HR departments: an assistant that screens applications quickly falls into a higher class and needs oversight.
- Banks and insurers: systems around credit and applications sit under strict scrutiny.
- Medical technology: AI in or near products faces extra conformity rules.
- Education providers: tools that grade performance touch high-risk questions.
Typical Use Cases
- Classify a company brain that only returns internal knowledge.
- Give a chat assistant a clear AI label.
- Review an HR-adjacent agent before it does pre-selection.
- Check the provider papers of a bought-in AI tool for duties.
- Build an internal register of all AI systems in use.
Decision Guide
| Signal | EU AI Act is central | Alternative |
|---|---|---|
| Decisions about people | full review needed | pure info lookup is lighter |
| Only retrieving internal knowledge | transparency and rights often suffice | no high-risk load |
| You buy a third-party system | check provider duties | clarify your deployer role |
| Pure test field with no effect | rather secondary | still respect data protection |
Example
A mid-sized firm wants an assistant for HR. Instead of letting it grade applications directly, the team first sets it up as pure research help, labels it as AI and rules that every recommendation is checked by a human. This describes the method, not legal advice and not an asserted metric.
Service Connection
We build an AI assistant or a company brain with labelling, rights and human approval, and set up AI agents so the right risk level is visible from the start.
Sources
This framing is a general description, not legal advice. The exact scope, deadlines and duties of the EU AI Act must be checked against the current regulation text and official guidance case by case.
