Yes. Internal use is still use. A business that runs an agent for its own staff is a deployer under the EU AI Act, and if it built the agent itself and put it into service, it is a provider too. The only carve-out for users is purely personal, non-professional activity, which no company qualifies for.
Why does internal use not take an agent out of scope?
The Act attaches duties to placing a system on the market and to putting it into service. Putting into service includes supplying a system for first use by the business itself. So an agent your engineering team assembles for the finance department has been put into service, and your company is its provider.
The deployer definition is equally broad: anyone using an AI system under their authority in a professional capacity. Your finance team using the agent makes your company its deployer. A company can hold both roles for the same agent. The basics of the regime are summarised in what the EU AI Act is, and the general test is in whether the EU AI Act applies to AI agents.
Which internal agents carry real obligations?
Most internal agents are minimal risk. A few are not.
- Agents used on staff. Annex III also lists certain employment uses as high-risk, in substance: recruitment and selection, and decisions affecting promotion, termination, task allocation and the monitoring or evaluation of workers. An internal agent that screens CVs or scores staff performance is high-risk, even though no customer ever sees it.
- Agents whose output reaches customers indirectly. An internal agent that pre-assesses loan applications for credit officers is doing creditworthiness assessment of natural persons. Being "back office" does not change the use.
- Agents staff talk to. Article 50 transparency duties are aimed at people interacting with AI systems, and employees are people. Where it is obvious to staff that they are using an AI tool, little more is needed; where it is not, tell them.
- All agents. Article 4 AI literacy applies to every system you put into use.
For high-risk systems used in the workplace, the deployer must inform workers' representatives and affected workers before putting the system into use.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
What may an internal agent do, and what stays with a person?
| Internal use | Agent's part | Stays with a person |
|---|---|---|
| Summarising policies for staff | Answers with citations | Policy interpretation disputes |
| Reconciling ledgers | Matches and posts within limits | Postings above the limit, write-offs |
| Screening job applications | Extracts and summarises; no ranking or rejection | Shortlisting and every rejection |
| Drafting staff performance notes | Collates facts from systems | Any rating or conclusion |
| Triaging IT tickets | Routes and resolves routine ones | Access grants to sensitive systems |
The HR rows are where internal agents most often drift into high-risk territory without anyone deciding they should.
An illustrative example
Take a mid-sized asset manager with an internal agent for its people team. It started as a tool to answer staff questions about leave policy. Six months later, a recruiter asked it to "rank these 40 applicants against the job description", and it did. Nobody had changed the agent's brief; it simply had access to the applicant folder.
At that moment, the firm was using an AI system for selection of candidates, a high-risk use, without the oversight, logs, literacy measures or worker information the Act expects. The fix was to remove the agent's access to recruitment files, add a refusal for ranking or scoring people to its brief, and route any future recruitment use through a separate, properly governed capability.
Which controls keep internal agents in order?
- An inventory of internal agents, with each capability and its use classified, kept current.
- Scope bounds that refuse uses the agent was not classified for, so a high-risk use cannot appear by accident.
- Source permissions limited to what the job needs; access to HR and applicant data is what turns a helper into a high-risk system.
- Role clarity: record where you are provider, deployer or both.
- A short notice to staff where it is not obvious they are dealing with an AI system.
Customer-facing agents raise different questions, covered in the EU AI Act for AI agents in customer service. The wider control map is in AI agent regulation, mapped to controls.
Where Pryme fits
In Pryme, an agent's role scope is one of five bounds checked at the moment of action. If an internal assistant is asked to rank applicants and that capability is not in its role, the action is refused and the verdict, naming the role bound, is filed.
Questions people ask
Does an internal pilot count?
A pilot used on real work by real staff is in use. Testing in a controlled setting before use is treated differently, but once outputs inform real decisions, treat it as deployed.
If we buy the agent from a vendor, are we still a provider?
Usually not; the vendor is the provider and you are the deployer. You can become a provider if you put your name on it, substantially modify it, or change its purpose to a high-risk one.
Do internal agents need a fundamental rights impact assessment?
Article 27 applies to certain deployers, including public bodies and those using high-risk systems for credit scoring or life and health insurance pricing. Most internal agents do not trigger it, but an internal credit agent can.