Buying a vendor's AI agent makes you its deployer under the EU AI Act. Before use, check how the agent is classified for your intended use and, if high-risk, that the provider has done its part. Oversight, monitoring, logs, notices, impact assessments and AI literacy stay with you, and changing the agent's purpose can make you the provider.
What does the Act require of buyers, in plain words?
The EU AI Act (Regulation (EU) 2024/1689) entered into force on 1 August 2024 and applies in phases. It splits obligations between providers, who develop a system or place it on the market under their name, and deployers, who use it under their authority.
For a high-risk system, the provider must meet the requirements in Articles 9 to 15, complete the conformity assessment the Act requires, draw up a declaration of conformity, affix the CE marking, register the system where required and give deployers instructions for use. The deployer, under Article 26, must use the system according to those instructions, assign human oversight to competent people, ensure input data it controls is relevant, monitor operation, keep logs under its control, inform workers and affected people where required, and report risks and serious incidents. Certain deployers must also carry out a fundamental rights impact assessment under Article 27. Article 50 transparency and Article 4 literacy apply whatever the tier.
Who does this apply to?
Any organisation in the EU, or whose agent's output is used in the EU, that uses a third-party agent professionally. A bank licensing a credit assessment agent, an insurer subscribing to a claims platform and a retailer deploying a vendor's service agent are all deployers.
Buying an agent also creates third-party risk duties outside the Act. Financial entities will meet DORA, under which agent providers become your ICT third parties and the vendor is very likely an ICT third-party provider, as explained in are AI agent vendors ICT third-party providers under DORA. US-regulated groups will recognise the theme from OCC third-party risk and AI agent vendors.
What should you check before using a bought agent?
Start with your intended use, not the vendor's brochure. An agent sold as a general assistant becomes high-risk if you use it to assess individuals' creditworthiness.
Then check, for any capability in a high-risk use:
- The provider has identified that use as intended and completed conformity work for it.
- Instructions for use describe the agent's purpose, capabilities, limits, accuracy, human oversight measures and logging.
- The agent automatically logs events, and you can access and keep the logs generated under your control.
- Oversight is designed in: you can hold actions for a person, override outputs and stop the agent safely.
- The provider will inform you of changes and of risks it discovers, and accept your findings.
For agents not in a high-risk use, check transparency: will the agent tell people they are dealing with AI, and does it mark generated content? A general method for vendor review is set out in NIST AI RMF third-party risk for AI agent platforms.
When does the buyer become the provider?
In substance, a deployer takes on provider duties if it puts its own name or trademark on a high-risk system, makes a substantial modification to one, or changes the intended purpose of a system so that it becomes high-risk.
Agents make the last two easy to trigger. Configurable agents let you add sources, tools and instructions. Giving a general service agent access to bureau data and asking it to recommend lending decisions changes its intended purpose to a high-risk one. At that point you, not the vendor, carry the provider's obligations for that use. Capability boundaries enforced by the system, and a change process that checks classification, are the practical defence.
How do you meet the deployer duties for a bought agent?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Use according to instructions | Agent capabilities limited to the provider's intended use | Capability configuration matched to instructions |
| Classification per use | Each capability checked for high-risk use before activation | Classification record |
| Human oversight by competent people | Named overseers trained on the agent's limits | Assignments and training records |
| Input data relevance | Sources approved per capability | Source register |
| Monitoring and provider feedback | Behaviour reviewed; findings sent to the provider | Monitoring reports and provider notices |
| Logs kept | Action record retained and exportable | Retention settings and export tests |
| Transparency and notices | Disclosure at first contact; decision notices where required | Conversation and notice records |
| Impact assessment | FRIA where Article 27 applies, alongside the DPIA | Signed assessments |
| Change does not make you provider | Configuration changes reviewed for purpose and substantial modification | Change records with classification check |
An agent can help with vendor review work, as discussed in can an AI agent assess third-party risk, but acceptance stays with people.
What will a supervisor ask?
- For this agent, are you the deployer or the provider, and why?
- Show us the provider's instructions for use and how your configuration matches them.
- Who oversees the agent, and what training did they have?
- What findings have you sent to the provider?
- Have you changed the agent's sources or tools since you bought it? Did that change its purpose?
What does the Act not cover for bought agents?
It does not set commercial terms, so contract rights to logs, change notice and information depend on negotiation. It does not replace sector third-party rules such as DORA. It does not transfer deployer duties to the vendor however the contract is written. For the full map, see AI agent regulation mapped to controls.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Questions people ask
Does a vendor's CE marking mean we have no work to do?
No. It covers the provider's obligations. Deployer duties on use, oversight, monitoring, logs and notices remain yours.
Is adding our own documents to a bought agent a substantial modification?
Not necessarily. Adding sources within the intended purpose is usually use, not modification. Using them to take the agent into a new, high-risk purpose is the risk.
Can we use an agent whose provider is outside the EU?
Yes, if the provider meets its obligations, including appointing an authorised representative in the EU where required for high-risk systems.