EU Commission Publishes AI Act Transparency Guidelines

What the Commission published

The European Commission published guidelines on 20 July 2026 to help providers and deployers of certain AI systems prepare for transparency obligations under the EU AI Act. The Commission says the obligations start to apply on 2 August 2026. The guidance is not a substitute for the regulation itself or for legal advice, but it gives organizations a practical reference for understanding which systems and interactions may require notices, labels, or other information for people affected by AI.

Transparency is not just a website notice. In practice, it concerns whether people can understand when they are interacting with an AI system, when content has been generated or manipulated by AI, and what information a provider or deployer must make available in the relevant context. The correct obligation depends on the system, the role an organization plays, the intended use, and the audience.

Start with an inventory, not a generic policy

Organizations should first map their AI uses: customer-facing chat, internal decision support, content-generation tools, biometric or emotion-related functions, and vendor products embedded in a wider workflow. For each use, record the provider, deployer, model or system version, data flows, user groups, jurisdictions, and the point at which an individual encounters the output. This makes it possible to determine which legal duties and product controls are actually relevant.

A broad statement that a company uses AI is rarely sufficient. Notices must be understandable, timely, and appropriate to the interaction. Teams should test whether a person can see the information before relying on the system, whether the language fits the audience, and whether the notice remains correct after a model, interface, or workflow changes.

Build operational evidence

Compliance work needs owners and evidence. Keep versioned records of risk assessments, user notices, labeling decisions, vendor assurances, test results, and decisions about exceptions. Where a supplier provides a system, clarify in writing who supplies technical documentation, who controls changes, and how information required for downstream transparency will be delivered.

Human review remains important for high-impact uses. A label does not make an unreliable output safe, and disclosure does not remove obligations around privacy, security, discrimination, consumer protection, or sector rules. Teams should define escalation routes, monitor complaints and failures, and retain a way to pause or alter a deployment.

A practical next step

Before 2 August, choose the most exposed workflows and run a short readiness review. Confirm the legal role of each party, compare current product behavior with the guidance, test the user journey, and assign a named owner for ongoing monitoring. The aim is not paperwork for its own sake; it is to give people meaningful information and to make AI use accountable as systems evolve.