The part of the AI Act that is live now
Most commentary on the Digital Omnibus focused on what was delayed. Article 50 was not. Since 2 August 2026, providers and deployers in scope have had transparency duties: people should know when they are interacting with an AI system, generative output should be marked in a machine-readable way, and deepfakes and certain AI-generated public interest text should be disclosed.
What the Code of Practice adds
On 10 June 2026 the European Commission published the final Code of Practice on transparency of AI-generated content. It has two parts. The first is for providers of generative AI systems and deals with marking and detection. The second is for deployers who publish deepfakes or AI-generated text intended to inform the public on matters of public interest, and deals with labelling.
Adhering to the Code is voluntary, and it is not conclusive proof of compliance. The underlying Article 50 duties are binding either way. Commentators note the final version made several previously mandatory measures optional, introduced an EU icon for labelling, and set an interoperability deadline of 2 February 2027 for marking approaches.
Where product teams get caught out
The legal text is short. The product decisions behind it are not. A few that tend to surface late:
- Who counts as the provider when you build on a third-party model and add your own generation feature on top.
- Whether marking survives your own pipeline. Resizing, compression and format conversion can strip metadata that a model provider added upstream.
- Where the disclosure appears for a conversational agent that hands off between AI and human staff.
- Whether a piece of AI-assisted text has had enough human review and editorial responsibility to fall outside the labelling duty, and who decides that.
A practical starting point
List every surface where your product generates or manipulates text, images, audio or video. For each, record the role you play, whether output is marked, whether that marking survives to the user, and where a person would see a disclosure. That inventory is usually the first time a team sees the whole picture in one place.
We cover transparency alongside the wider risk picture in the AI Opportunity and Governance Sprint.