AI Act transparency starts on 2 August 2026

Why this matters now

The European Commission says AI Act enforcement and new transparency requirements start on 2 August 2026. For builders, this turns Article 50 from a planning item into an operating checklist. It does not mean every AI use is high-risk. It means certain user-facing AI interactions and generated-content surfaces now need visible, testable, and recorded transparency controls.

What Article 50 covers

• Tell users when they are directly interacting with an AI system, unless this is obvious from context.
• Mark AI-generated or manipulated audio, image, video, and text outputs where provider output-marking duties apply.
• Disclose deepfakes and certain AI-generated public-interest text.
• Separate provider duties from deployer duties instead of treating every party as having the same obligation.
• For public-interest text, keep human review and editorial responsibility clear where they are part of the publication workflow.

The practical Burhuc interpretation

Burhuc already marks public publications as human reviewed with Burhuc editorial responsibility. We are keeping that rule because Article 50 is partly about whether people can tell when content is AI-generated or manipulated, and whether human review and editorial control are part of the publication path. The disclosure has to be close enough to the experience to be useful, not buried in a policy page.

What builders should check this week

• Chatbots and agents: the first screen should tell users they are interacting with AI before meaningful use.
• Generated guidance or remediation output: show AI origin, provider or model metadata, creation time, and review warnings where appropriate.
• Marketing and editorial text: record whether AI assisted the draft, who reviewed it, and who accepted editorial responsibility.
• Images, video, and audio: add labels and provenance where content is synthetic or materially altered.
• Vendor map: distinguish foundation model provider, system provider, deployer, customer, and end user roles.
• Evidence: keep screenshots, DOM attributes, JSON-LD or metadata, policy text, and review logs.

How this changes Burhuc products

• CertentiScan now gains an automated AI-transparency signal for visible AI assistant or generated-content surfaces.
• CertentIQ keeps provider, model, disclosure, and review metadata as part of agent evidence.
• CertentiTrain treats disclosure and human-review wording as remediation material when AI-facing surfaces are evaluated.
• CertentiWall will be developed as a reusable disclosure and provenance layer for AI-facing workflows.

What this does not prove

A visible AI label is not full AI Act compliance. High-risk classification, GDPR, contracts, security controls, logging, provider obligations, and sector-specific duties may still require separate review. The useful move is to make transparency visible, repeatable, and evidenced before the product is under pressure.

Official sources: Open Commission release - European Commission notice on AI Act enforcement and transparency requirements from 2 August.

Open EUR-Lex AI Act - Regulation (EU) 2024/1689, Artificial Intelligence Act.

Open Article 50 - AI Act Service Desk text for Article 50 transparency obligations.

Open transparency FAQ - European Commission FAQ on Article 50 transparency obligations.

Open enforcement page - European Commission AI Act enforcement information.