Mostly to use agents responsibly and prove it. For high-risk agents, Article 26 requires deployers to follow the provider's instructions, assign human oversight to competent people, monitor operation, keep the logs under their control, report serious incidents and inform affected people and workers. Every deployer, whatever the risk tier, must ensure AI literacy and meet transparency duties.
What are the deployer's duties, in substance?
A deployer is any person or organisation using an AI system under its authority in a professional capacity. The duties depend on the agent's tier.
For every agent: - AI literacy (Article 4): staff dealing with the agent have a sufficient understanding for their role. - Transparency (Article 50): where deployers use emotion recognition or create certain synthetic content, people must be informed; and the provider's design should make sure people know they are interacting with an AI system. - No prohibited practices: these have applied since 2 February 2025.
For high-risk agents (Article 26), in addition: - Use the agent in line with the provider's instructions for use. - Assign human oversight to people with the necessary competence, training, authority and support. - Ensure input data under the deployer's control is relevant and sufficiently representative for the purpose. - Monitor operation; if a risk appears, inform the provider or distributor and the market surveillance authority, and suspend use where needed; report serious incidents. - Keep the automatically generated logs under the deployer's control for an appropriate period, at least as long as the Act's minimum. - Inform workers' representatives and affected workers before using a high-risk system in the workplace. - Inform individuals when a high-risk system makes or assists decisions about them. - Use the provider's information to carry out a data protection impact assessment where required.
Some deployers must also complete a fundamental rights impact assessment under Article 27: public bodies and private bodies providing public services, and deployers assessing creditworthiness or pricing life and health insurance.
A plain summary of the Act is in what the EU AI Act is, and the order in which these apply is in the EU AI Act timeline for AI agent deployers.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Why are these duties harder when the system acts?
The Article 26 list assumes a person sits between the AI system's output and its effect. With an agent, the output is often the effect. Three duties change shape as a result.
Human oversight must work before the action, so held actions wait for a person rather than being reviewed afterwards. Monitoring must look at what the agent did, not only whether it ran. Logs must record actions and their reasons in enough detail to reconstruct a decision; output logs alone fall short, as discussed in whether Article 12 requires AI agents to keep logs. The wider argument is in the EU AI Act and agents that act.
What may the agent do, and what stays with a person?
The agent may gather data, assess and recommend, and act within limits on routine, reversible work. A named overseer keeps decisions with significant effects on individuals, the authority to pause or stop the agent, the judgement that a risk or serious incident has arisen, and any decision to use the agent outside its intended purpose.
How do the duties map to controls?
| Deployer duty | Control for an agent | Evidence |
|---|---|---|
| Use per instructions | Agent's scope and limits match the provider's intended purpose | Configuration record |
| Competent oversight | Named overseers, trained on this agent, with authority to stop it | Training and authority records |
| Relevant input data | Sources approved per capability | Source register |
| Monitoring and incidents | Outcome monitoring; automatic hold on anomalies | Alerts, holds, incident reports |
| Logs | Per-action record with sources | Log store under deployer control |
| Inform people | Notices before or at the decision | Notice templates and delivery records |
An illustrative example
Take a car finance company in Italy deploying a platform's credit assessment agent. It reads the provider's instructions for use and finds the agent is intended for consumer loans up to a stated amount, so it blocks the agent from larger applications. It names four credit analysts as overseers, trains them on the agent's documented weaknesses, and gives them authority to pause it. Logs are exported daily to the company's own store. Applicants are told an AI system assists the assessment. Before go-live, it completes the Article 27 assessment. When a spike in referrals appears one week, the agent is paused, the provider is informed, and the cause, a changed bank statement format, is fixed before use resumes.
Questions people ask
What if we change the agent beyond the provider's instructions?
Substantially modifying a high-risk system, or using it for a high-risk purpose the provider did not intend, can make you a provider, with the provider's much heavier duties.
What are the penalties for deployers?
The Act sets fines that scale with the type of breach and the size of the company. The detail is in EU AI Act penalties and what they mean for agent deployers.
Do these duties apply to agents we only use internally?
Yes. Internal use is use. High-risk uses such as staff evaluation or credit assessment trigger the same duties. The full map is in AI agent regulation, mapped to controls.