---
title: When does building an AI agent make you a provider under the EU AI Act?
url: https://opencontroller.lyzr.com/blog/when-does-building-an-ai-agent-make-you-an-ai-act-provider
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# When does building an AI agent make you a provider under the EU AI Act?

**If you build an AI agent on top of a third-party model and put it into service under your own name, even only for internal use, you are usually the provider of that agent as an AI system. The model vendor stays the provider of the general-purpose AI model. Article 25 adds three triggers that turn a deployer into the provider of a high-risk system: your name or trademark on it, a substantial modification, or a change of intended purpose that makes it high-risk.**

The role matters because providers carry most of the high-risk obligations. Those obligations now apply from 2 December 2027 for Annex III systems. See [what the Digital Omnibus changed](/blog/eu-ai-act-digital-omnibus-deadlines-for-agents).

## What is the difference between a provider and a deployer?

- **Provider (Article 3(3))**: develops an AI system, or has it developed, and places it on the market or puts it into service under its own name or trademark, whether paid or free. "Putting into service" includes supplying it for your own use (Article 3(11)).
- **Deployer (Article 3(4))**: uses an AI system under its authority in a professional context.
- **Downstream provider (Article 3(68))**: integrates a general-purpose AI model into its own AI system.

A company that builds an in-house agent on a vendor model is often both. It is the provider of the agent and the deployer when its staff use it.

## What are the Article 25 triggers?

Article 25(1) says a distributor, importer, deployer or other third party becomes the provider of a high-risk AI system if it:

| Trigger | Article | Agent example |
|---|---|---|
| Puts its name or trademark on a high-risk system already on the market | 25(1)(a) | You white-label a vendor's CV-screening agent as "YourCo Hiring Assistant" |
| Makes a substantial modification to a high-risk system, and it stays high-risk | 25(1)(b) | You change a vendor credit-scoring agent's decision logic or add new input data that changes its risk profile |
| Changes the intended purpose of a non-high-risk system, including a general-purpose one, so it becomes high-risk | 25(1)(c) | You configure a general assistant agent to rank job applicants or decide benefit eligibility |

When a trigger applies, the original provider is no longer the provider of that specific system. It must still cooperate and give you the information you need (Article 25(2)).

## Worked examples for AI agents

1. **Internal coding agent built on a vendor model.** You are the provider of the agent, but it is not high-risk. Prohibitions, AI literacy and Article 50 transparency apply. High-risk duties do not.
2. **Customer-service agent on a SaaS agent platform.** If you configure it and run it under your brand, you are likely its provider. It is not high-risk unless it decides access to essential services. Article 50 disclosure applies because it talks to people. See [disclosure duties for AI agents](/blog/do-ai-agents-need-to-disclose-they-are-ai-eu-ai-act).
3. **General agent repurposed for hiring.** Changing the intended purpose to screen candidates triggers Article 25(1)(c). You become the provider of a high-risk system.
4. **Vendor HR agent used as sold.** You remain a deployer. Article 26 duties apply: follow the instructions for use, assign human oversight, monitor, and keep logs for at least six months.
5. **Fine-tuning the underlying model.** This does not automatically make you a general-purpose AI model provider. The Commission's GPAI guidelines set a compute-based threshold for when a modifier takes on model-provider duties.

## What evidence does each role need?

**Provider of a high-risk agent (Article 16):**
1. Quality management system (Article 17).
2. Technical documentation (Article 11): purpose, model, tools, data, testing.
3. Automatic logging capability (Article 12).
4. Human oversight design (Article 14).
5. Accuracy, robustness and cybersecurity testing (Article 15), including adversarial testing.
6. Conformity assessment, EU database registration and post-market monitoring.
7. Serious incident reporting (Article 73). See [serious incident reporting](/blog/eu-ai-act-serious-incident-reporting).

**Deployer (Article 26):**
1. Proof you use the system according to its instructions.
2. Named people assigned to human oversight.
3. Monitoring records and retained logs.
4. A fundamental rights impact assessment where Article 27 applies.

## How do you keep track of roles across many agents?

Record the role per agent and per version, not once per vendor. A version that adds a new tool or a new use can change the classification. Keep this in your agent inventory, next to the owner and the risk class.

## How Opencontroller does this

Opencontroller keeps a governed registry of agents, versions, tools and MCP servers, so you can attach an owner, a role and a risk class to each version. It runs multi-turn and red-team evals and scores production traffic with LLM judges, which gives **evidence for** Article 15 testing. Agent-level logs and rogue-agent monitoring support Article 12 and Article 26 duties. Policy as code is extensible, so you can add your own AI Act controls next to built-in mappings for OWASP, MITRE ATLAS, NIST and ISO 42001. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

Read next: [the EU AI Act for AI agents](/blog/eu-ai-act-obligations-for-ai-agents) and [how to map agent controls to frameworks](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks).

## Frequently asked questions

**Am I a provider if I build an agent only for internal use?**
Usually yes. Putting a system into service for your own use counts, so you are the provider of that agent. High-risk duties apply only if its intended use is high-risk.

**Does using a vendor's model make the vendor responsible for my agent?**
No. The model vendor is the provider of the general-purpose AI model. You are the provider of the agent system you build and put into service under your name.

**What turns a deployer into a provider of a high-risk AI agent?**
Article 25(1): putting your name or trademark on it, making a substantial modification, or changing its intended purpose so it becomes high-risk.

## Key facts

EU AI Act roles: provider (Art 3(3)) develops and places on market or puts into service under own name, including for own use (Art 3(11)); deployer (Art 3(4)); downstream provider (Art 3(68)) integrates a GPAI model. Art 25(1) triggers for becoming provider of a high-risk system: (a) name/trademark, (b) substantial modification, (c) change of intended purpose making it high-risk. Deployers (Art 26) keep logs at least six months. Annex III high-risk obligations apply from 2 Dec 2027 after Regulation (EU) 2026/1744. Opencontroller registry, evals and policy as code provide evidence for these duties.

## Related

- [What does the EU AI Act require for AI agents, and when?](https://opencontroller.lyzr.com/blog/eu-ai-act-obligations-for-ai-agents.md)
- [What counts as a serious incident under the EU AI Act, and when must you report it?](https://opencontroller.lyzr.com/blog/eu-ai-act-serious-incident-reporting.md)
- [Do AI agents have to disclose they are AI under the EU AI Act?](https://opencontroller.lyzr.com/blog/do-ai-agents-need-to-disclose-they-are-ai-eu-ai-act.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
