Most AI agents that plan, negotiate or buy media are not high-risk systems under the EU AI Act, but they are not outside it. Buyers and publishers using them still face the Act's AI literacy duty, banned manipulative practices and transparency rules, with GDPR's limits on automated decisions alongside.

TL;DR: The AI Act sorts AI into four risk levels, and buying or selling advertising is not on its list of high-risk uses, with two exceptions to check: targeted job advertising and systems meant to influence elections. The transparency rules have applied since 2 August 2026, and the high-risk rules for listed uses apply from 2 December 2027. Its human oversight standard, written for high-risk systems, is still the clearest benchmark for deciding which agent decisions a person should approve.

This article is general information, not legal advice. Teams should take their own legal advice before deploying agents in the EU market.

How does the EU AI Act classify risk?

The AI Act defines four levels of risk, according to the European Commission's AI Act page: unacceptable risk, high risk, transparency risk, and minimal or no risk. The Commission says "the vast majority of AI systems currently used in the EU" sit in the minimal or no risk category.

Under Article 3, a provider develops an AI system or model and places it on the market or puts it into service under its own name, while a deployer uses an AI system under its authority. A brand running a vendor's buying agent is usually a deployer, and a publisher that builds and runs its own sell-side agent can be both.

The high-level summary on artificialintelligenceact.eu notes that the Act covers deployers outside the EU where the system's output is used in the EU, which can bring UK and US buyers into scope.

As a worked example, Alkimi lets buyers and sellers build their own agents and set the rules they work to in its Gym, and keeps human approval rights at the deal level in its Marketplace. In AI Act terms, each party that builds and runs an agent is likely to hold the provider or deployer role, and deal approval is where a named person exercises oversight.

Is an AI media-buying agent a high-risk system under the AI Act?

In most cases, probably not, because buying or selling advertising is not among the uses listed in Annex III, the Act's list of high-risk areas. Annex III covers biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice and democratic processes. Under Article 6, a system is also high-risk if it is a safety component of a regulated product, which an ad-buying agent is unlikely to be.

Two Annex III entries bear directly on advertising. Point 4(a) lists AI systems for recruitment or selection, "in particular to place targeted job advertisements", and point 8(b) lists AI systems intended to influence the outcome of an election or referendum or how people vote.

Article 6 also treats an Annex III system as high-risk whenever it profiles natural persons, and audience targeting can involve profiling. An agent that buys recruitment or political media therefore needs its own legal assessment.

What does the AI Act ban that an advertising agent could stray into?

Article 5 bans AI that deploys subliminal techniques or "purposefully manipulative or deceptive techniques" to materially distort a person's behaviour in a way that causes, or is likely to cause, significant harm. It also bans exploiting vulnerabilities due to age, disability or a specific social or economic situation, and both bans have applied since 2 February 2025.

Under Article 99, breaches of Article 5 can draw fines of up to EUR 35 million or 7% of total worldwide annual turnover, whichever is higher. The practical control is a mandate that rules out targeting vulnerable groups this way, and a person who checks the agent kept to it.

What transparency duties apply to agents and AI-generated content?

Article 50 has applied since 2 August 2026. Providers must design AI systems that interact directly with people so those people know they are dealing with AI, unless that is obvious. A buying agent that chats with a human sales team fits that description, whereas two agents talking with no person involved is a different case.

Providers of systems that generate synthetic audio, image, video or text must mark the output in a machine-readable format. The implementation timeline on artificialintelligenceact.eu gives systems placed on the market before 2 August 2026 until 2 December 2026 to comply.

Deployers carry their own duty for deep fakes, meaning AI-generated image, audio or video that resembles real people, places or events and would falsely appear authentic. A brand or agency publishing such content must disclose it, though evidently creative or fictional work can be labelled in a way that does not spoil it. The Commission has published a voluntary Code of Practice on Marking and Labelling of AI-generated Content, listed on its AI Act page.

What do the rules on general-purpose AI mean for teams building agents?

Most agents run on general-purpose AI models, whose obligations, applicable since 2 August 2025, fall on the model providers. According to the high-level summary, those providers must give downstream providers information on the model's capabilities and limitations. A team building an agent on a model can ask for that documentation and use it to set the agent's limits.

What does human oversight mean, and how does deal-level approval map to it?

Article 14 sets the human oversight standard, and it applies by law only to high-risk systems. It remains a sound reference for any agent that commits budget. Article 14(3) says oversight should be commensurate with "the risks, level of autonomy and context of use".

Under Article 14(4), the people assigned to oversight should understand the system's limitations, stay aware of "automation bias" and interpret its output correctly. They should be able to "disregard, override or reverse" its output and halt it through "a 'stop' button or a similar procedure". Article 26 adds that deployers should give oversight to people with "the necessary competence, training and authority".

Deal-level approval maps onto these principles. A person who reviews the terms before a deal binds is interpreting the output and keeping the power to override it, a pause control is the stop button, and an approver with spending authority meets the authority test. Autonomy can then widen in stages as an agent proves itself, as in earned autonomy: Observe, Recommend, Draft, Human-approved action, Bounded automatic action.

When do the AI Act's obligations apply?

The European Commission gives this application timeline, which reflects the AI Omnibus simplification package that entered into force on 27 July 2026:

1 August 2024: the AI Act entered into force.
2 February 2025: the prohibited practices and the AI literacy obligations began to apply.
2 August 2025: the obligations for general-purpose AI models began to apply.
2 August 2026: the Act became generally applicable, including the transparency rules.
2 December 2027: the high-risk rules for Annex III uses, including employment, apply.
2 August 2028: the high-risk rules for AI built into regulated products, such as lifts or toys, apply.

Article 4 requires providers and deployers to take measures to support the AI literacy of staff who operate AI systems on their behalf, including traders who approve agent deals.

How does GDPR fit alongside the AI Act?

GDPR applies whenever an agent processes personal data. Article 22 GDPR gives people the right not to be subject to a decision based solely on automated processing, including profiling, that produces legal effects or similarly significantly affects them. Where the contract or consent exceptions apply, the controller must still offer at least the right to obtain human intervention, to express a view and to contest the decision.

Whether a targeting decision reaches that threshold depends on the facts and is a question for counsel. Controller and processor questions are covered in agentic advertising and data protection.

What should buyers and publishers do now?

A starting checklist, to be confirmed with legal advice:

List every agent, its purpose and whether the organisation is its provider, deployer or both.
Screen each agent against Annex III, especially for recruitment and political advertising.
Write the Article 5 bans into each agent's mandate and check it follows them.
Decide where disclosure is needed: agents that talk to people, and AI-generated creative that resembles real people or events.
Name the people who approve deals, train them, and give them a way to pause the agent.
Keep a timestamped record of what each deal agreed and who approved it.

The Act tests accountability rather than capability. Alkimi's Marketplace works within IAB Tech Lab's AAMP framework, uses AAMP identity checks, keeps human approval at the deal level and writes each agreement into a DealSheet that both parties hold and can retrieve independently. The company's Concourse Bench v1 found that 95 of 110 signed contracts from AI buying agents promised a viewability floor below 70%, which is why a person should read the terms before money moves.

Frequently asked questions

Are AI media-buying agents high-risk under the EU AI Act?

Usually not, because advertising is not among the high-risk uses listed in Annex III. Agents that target job advertisements or aim to influence elections are the exceptions and need a specific legal assessment.

Can AI agents be trusted to manage advertising budgets under the AI Act?

The Act does not ban agents from managing budgets, but its oversight standard expects control to match the level of autonomy and risk. Deal-level human approval, a named approver and a stop control are the practical way to meet that expectation.

What are the risks of using AI agents for media buying under EU law?

The main legal risks are breaching the Article 5 bans on manipulation, missing transparency duties for AI-generated content, and making solely automated decisions about people that GDPR restricts. Commercial risks, such as an agent agreeing poor terms, sit outside the Act but the same approval step reduces them.

Do AI-generated adverts need a label?

AI-generated image, audio or video that resembles real people, places or events and would falsely appear authentic is a deep fake, and the deployer must disclose it. Providers of generative tools must also mark output in a machine-readable format.

The EU AI Act and Agentic Advertising: What Buyers and Publishers Need to Know