
- EU AI Act
- AI Act Complaints Tool
- European AI Office
- European Data Protection Supervisor
- Market Surveillance Authorities
- AI Governance
In this article (4)
AI Act Complaints Tool turns complaint intake into product governance
Key Takeaways
- Map each complaint type to the AI Office, EDPS or national authority before users force the question.
- Treat complaint intake as product governance, with records tied to system versions, suppliers and deployment context.
- Update vendor contracts so documentation and technical evidence can be produced when authorities ask.
- EU AI Act
- AI Act Complaints Tool
- European AI Office
- European Data Protection Supervisor
- Market Surveillance Authorities
- AI Governance
Why it matters
- ProductProduct leaders need complaint workflows that preserve technical context, not just customer support notes.
- InvestorsInvestors should ask whether portfolio companies can evidence AI Act compliance when complaints trigger scrutiny.
The useful compliance signal is not the form itself. It is the operating model it forces teams to build around evidence, routing and escalation.
Complaint mechanisms are where compliance slogans go to acquire timestamps. For AI teams in Europe, the useful signal is not another policy page promising supervision. It is that alleged infringements now have places to land, and those places are not all run by the same authority.
A Formiventos post says the European Commission introduced the AIActComplaintsTool, allowing individuals and organisations to submit complaints to the AI Office when they believe a provider or deployer may have infringed the AI Act. Treat that as a product signal, not a legal footnote. If a complaint can arrive through an official channel, your internal evidence trail needs to exist before the email from counsel does.
Three doors, not one inbox
According to the European Commission’s governance page, AI Act enforcement is shared among the European Commission’s AI Office, the European Data Protection Supervisor and national competent authorities designated by Member States. The EDPS says it supervises AI Act compliance for AI systems used by EU institutions, bodies, offices and agencies, and that natural or legal persons may submit complaints to it about alleged infringements involving those systems. Formiventos describes a separate AIActComplaintsTool route for complaints to the AI Office, tied to alleged infringements by providers or deployers of AI systems within the Office’s competence.
For planning purposes, the calendar matters because the enforcement machinery is no longer imaginary. The EDPS says the AI Act entered into force on 1 August 2024. The Commission’s governance page says Member States should have designated and empowered national competent authorities by 2 August 2025, and also refers to a July 2026 action plan on cybersecurity and AI.
AI Act governance dates to anchor
- 1 August 2024AI Act entered into force
- 2 August 2025national authorities
- July 2026cybersecurity and AI action plan
The practical translation is simple enough to be annoying. A complaint about an AI system used by an EU institution points toward the EDPS. A complaint about general-purpose AI model rules points toward the AI Office, according to the Commission’s market surveillance page. Complaints about AI systems in Member States generally belong with national market surveillance authorities, with a single point of contact required where a Member State appoints more than one authority.
What builders should wire into the product
The Commission’s market surveillance page says, in unusually plain terms, that “Complaints from individuals” will matter to effective AI Act enforcement. That sentence should make product teams look at their support and incident workflows with less romance. A user report alleging a prohibited practice, a high-risk system failure or a misleading output should not sit beside password reset tickets until someone remembers the legal department.
The same Commission page says market surveillance authorities can intervene when AI systems pose risks or fail to comply, conduct remote monitoring and access providers’ documentation, datasets and source code. That is the part builders should underline. Complaint intake is not just a form field; it is the front end of a possible documentation request.
In practice, this means AI governance needs named owners for intake, triage, evidence preservation and escalation. Product teams should be able to connect a complaint to the relevant system version, deployment context, model or supplier, user-facing decision and available documentation. Vendor contracts should require cooperation when documentation, datasets or source code access become relevant to an authority’s inquiry. We welcome clarity from regulators, as ever, means someone is asking where the logs are.
The split that will confuse users first
The Commission’s market surveillance page draws a line between the AI Office and national market surveillance authorities. The AI Office supervises and enforces rules for general-purpose AI models. National market surveillance authorities supervise and enforce rules for AI systems in Member States, including prohibitions and high-risk AI rules.
That split is tidy on paper and less tidy in products. A deployed AI service may combine a general-purpose AI model, a sector-specific workflow, a local deployer and a user interface that collects complaints. The person filing the complaint will not necessarily know which layer failed. The builder’s job is not to make the user solve the jurisdictional puzzle.
This is where a governance operating model earns its budget. Intake should capture enough context to route the issue without forcing complainants to classify the law. Escalation paths should distinguish model supplier issues, deployer configuration issues, EU institution use cases and Member State authority matters. If two authorities later disagree on scope, the team with a clean routing record will be less entertaining to read about.
What the law requires, and what it does not
EUR-Lex describes Regulation (EU) 2024/1689 as laying down harmonised AI rules for development, placing on the market, putting into service and use of AI systems in the Union. That does not mean every complaint proves an infringement. It does mean the operating environment now assumes complaints can become evidence.
The Commission’s governance page also notes that malfunctioning AI systems can affect rights such as privacy and non-discrimination. That is the bridge between product behavior and regulatory process. A broken classifier, opaque eligibility tool or poorly documented deployment is not merely a customer experience problem if it touches protected interests.
The useful move now is boring by design: map your complaint channels, preserve the records that explain what happened, and decide who can escalate to legal, security, model operations and vendor management. Do it before a regulator asks, not after. Complaint infrastructure is becoming part of AI product governance, which is another way of saying the paperwork is now connected to the build pipeline.
Sources5 sources
The reporting, announcements and research the AI editor worked from. Links open the original publisher.