EU high-risk AI guidance hits martech teams

Draft EU guidance sharpens the high-risk test for AI tools. Marketing and customer service teams need to classify systems and document compliance now.

Abstract violet flow-field artwork

Draft European Commission guidance clarifies which AI systems are high-risk under the EU AI Act. Marketing tools that assess people for jobs, credit, insurance, education or essential services need closer scrutiny. Teams should audit the stack, classify each tool against Annex III, and publish a public AI register before enforcement.

The legal trigger is Article 6(2) of Regulation (EU) 2024/1689. It states, "In addition to the high-risk AI systems referred to in paragraph 1, AI systems referred to in Annex III shall be considered high-risk." The Commission's 2025 Draft Commission Notice on the application of Article 6 of Regulation (EU) 2024/1689 to high-risk AI systems gives practical tests for applying that rule. Jones Day's client alert summarises the practitioner implications, but the Commission text sets the classification logic.

Why marketing tools can trigger Annex III

Most content generation, ad targeting, SEO testing and email optimisation tools remain outside the high-risk class. They do not decide a person's access to a job, credit, insurance or an essential service. The position changes when a tool's output moves into candidate screening, credit pre-qualification, insurance pricing or eligibility decisions.

Annex III point 4(a) gives the clearest example. It covers AI systems "intended to be used for the purposes of recruitment or selection of natural persons, notably to target job advertisements, screen or filter applications, or evaluate candidates in the course of interviews or tests." A marketing label does not remove a tool from that category if the actual use matches it.

Salesforce's Einstein Lead Scoring shows how the line moves. Used to rank leads for sales follow-up, the tool usually stays outside Annex III. If an integration feeds that score into credit pre-qualification or candidate shortlisting, the use becomes high-risk, even though the vendor sells it as a marketing product. The draft guidance pushes buyers to assess the function, the data and the decision influenced, not the tool's label.

Annex III covers employment, education, creditworthiness, insurance pricing, biometric identification and essential private and public services. The duty to assess falls on the deployer as well as the vendor. A marketing operations team that buys a general tool and connects it to a high-risk purpose takes on deployer responsibilities.

How strong is your own AI marketing? Take the free two minute scorecard and get an honest read plus the next steps that matter most.

Take the scorecard

For marketing operations leads

Start with a one-page AI register. List every tool across marketing, sales, HR and customer service. For each tool, record the declared purpose, the personal data processed, the decision it supports, the vendor's AI Act classification and the review frequency. Use one operational metric. Budget 60 minutes per tool for first-pass triage. That limit keeps the audit moving without skipping evidence.

Follow a B2B lead-scoring tool's output beyond marketing. If the score also feeds a candidate screening report or a credit pre-qualification form, classify the tool as high-risk until a legal review confirms otherwise.

Sample AI register row

Field Entry
Tool Lead scoring model connected to CRM
Declared purpose Rank marketing-qualified leads for sales follow-up
Data processed Contact firmographics, email engagement, campaign source
Decision it supports Lead routing and call prioritisation
Annex III trigger None while output stops at marketing routing
Review status Low-risk on file, reassess if output feeds credit or hiring screening
Owner Marketing operations manager
Review frequency Quarterly

High-risk triage table

Condition Answer Next action
Does the tool affect employment, credit, insurance, education, biometric identification or an essential service? Yes Treat as high-risk until legal review confirms the exception
Does the vendor rely on an Article 6 exception? Written analysis only Ask for the declared intended purpose and legal basis
Does the tool only create or optimise marketing content? Yes Document as out of scope and keep the evidence
Is the tool supplied by a third party but used for a high-risk purpose? Yes Record deployer responsibilities in the contract and assign an internal owner

Vendor questions to ask before signing

  • What intended purpose did you declare for this system under the EU AI Act?
  • Are you relying on an Annex III exception? Send the written analysis and Article 6 legal basis.
  • Does the model score, classify or rank individuals for employment, credit, insurance, education or essential services?
  • Which human oversight controls do you provide, and which controls must we add as the deployer?
  • Will you notify us if the system or its intended purpose changes in a way that alters the classification?

For compliance and legal teams

Compliance owns the classification file. The draft guidance treats the vendor's declared intended purpose as a starting point. Your actual use and foreseeable misuse can move a system into the high-risk class. Document both. If they differ, get a written legal opinion before launch.

Ask for the vendor's technical documentation and Article 6 analysis. Do not accept a one-line claim that a system is low-risk. For each Annex III category, record whether the tool influences a decision, whether it profiles or ranks individuals, and whether an exception applies. Keep that evidence in the same register.

Human oversight needs to be specific. Name the person or role that reviews high-risk outputs, state what they can override, and document how they access the information. That detail matters more than a general statement of intent.

For SEO and content leads

AI search tools respond to structured, attributable governance pages. When an assistant answers a question about the EU AI Act or high-risk classification, it cites sources that explain the rule and show how a business applies it. Publishing the public version of your AI register, triage process and vendor criteria gives the model something to quote.

Legal should own the register. Marketing should own the public explanation. Use headings, bullet points and FAQ schema. Add named owners, decision criteria and revision dates. This page should read as a working compliance document, not corporate storytelling.

Role-specific questions

What should a marketing operations lead do first?

Build the AI register. List every tool across marketing, sales, HR and customer service, then run the 60-minute triage test. If a tool touches employment, credit, insurance, education or essential services, treat it as high-risk until legal review confirms otherwise.

Is a chatbot that gives marketing support high-risk?

Not automatically. A chatbot that describes a credit card or product is usually low-risk. It becomes high-risk if it assesses eligibility for credit, insurance, employment or an essential service, or if it processes biometric data to identify a person.

When must the high-risk audit be complete?

For most Annex III systems, the main obligations apply from 2 August 2026. The register, triage and vendor reviews should be complete before that date because contract changes and tool replacements can take several quarters.

Put this to work on your business

We build exactly this, SEO, AI visibility, PPC, email and automation, as one measured system. Talk it through with a real person, no obligation.

Book a free strategy call or take the two minute scorecard