Short answer: On July 20, 2026 the European Commission adopted its final guidelines on the AI Act’s Article 50 transparency obligations, and confirmed the Code of Practice on Transparency of AI-generated Content as the recognized compliance framework. Article 50 obligations apply from August 2, 2026 — and unlike the high-risk obligations recently postponed under the Digital Omnibus, this deadline did not move. Teams that expose AI interaction or AI-generated content to EU users need their disclosure and marking story settled now.
What happened
Two related decisions landed in the final stretch before the deadline:
- Final Article 50 guidelines (July 20). The Commission adopted implementation guidelines for the transparency obligations that cover, among other things, AI systems interacting directly with people and AI-generated or manipulated content.
- Transparency Code of Practice confirmed. The Commission’s assessment recognized the Code of Practice on Transparency of AI-generated Content as an EU-wide framework for meeting marking and labelling obligations.
The context matters: in June, the Council gave final approval to the Digital Omnibus package, which postponed the AI Act’s high-risk system deadlines (Annex III to December 2027, Annex I to August 2028). Article 50 transparency obligations were not part of that postponement. August 2, 2026 stands.
Who in engineering is actually affected
Article 50 is not aimed only at consumer chatbots. Engineering organizations should map at least three exposure paths:
| Exposure | Typical example | Question to answer before August 2 |
|---|---|---|
| AI systems interacting with people | A support or coding assistant embedded in your product for EU users | Is the AI nature of the interaction disclosed where it is not obvious? |
| AI-generated content shown to users | Generated text, code explanations, images, or summaries surfaced in your product | Are marking/labelling obligations met per the recognized Code of Practice? |
| Internal tools reaching external users | An internal agent whose output ships into customer-facing surfaces | Does the output path carry the required transparency downstream? |
AI coding tools used purely internally are generally a governance rather than an Article 50 problem — but output that flows into user-facing products can change that analysis. This is a legal determination; route it through counsel with a concrete inventory of your AI touchpoints. Our EU AI Act timeline guide covers the broader schedule.
What to do this week
- Inventory AI touchpoints that face EU users: interactions, generated content, and any surfaces where AI output is republished.
- Read the final guidelines against your inventory — the July 20 text is the version that counts, not earlier drafts.
- Adopt or map to the Code of Practice for content marking; the Commission has confirmed it as the recognized EU-wide framework.
- Document decisions. Regulators ask for evidence trails, and so do enterprise customers. Keep the same discipline you would apply to a governance program for AI coding tools.
A 60-minute readiness pass
If August 2 is close and you have not started, this bounded pass gets a first draft of the evidence regulators and enterprise customers ask for. It is triage, not a substitute for legal review.
- List AI touchpoints (20 min). Enumerate every place your product interacts with EU users through AI or shows them AI-generated content. Include embedded assistants, generated summaries, and any internal agent whose output reaches customers.
- Classify each against Article 50 (15 min). For each touchpoint, note whether it is direct AI interaction, generated content, or out of scope, and why. Flag anything ambiguous for counsel rather than guessing.
- Check disclosure and marking (15 min). Where the AI nature is not obvious, confirm a disclosure exists; where content is generated, confirm marking maps to the recognized Code of Practice.
- Record the decisions (10 min). Capture the inventory, classifications, and gaps in one document with an owner and a date. That record is the artifact both regulators and enterprise buyers request.
The output is not a compliance certificate — it is a defensible starting point that turns a vague deadline into a short, owned punch list.
What this signals beyond August 2
Article 50 is the first hard AI Act deadline that did not slip, and that pattern is the real planning signal. The Digital Omnibus moved the high-risk classification dates precisely because they are complex and contested; transparency obligations, being more mechanical, stayed put. Expect the same logic on future dates: obligations that are concrete and low-ambiguity are the least likely to be postponed. Teams that build a durable AI-touchpoint inventory now — rather than a one-off scramble — will absorb the next deadline as an update to an existing document instead of a fresh fire drill.
The takeaway
The Digital Omnibus bought time for high-risk classification work — not for transparency. If your product interacts with EU users through AI or shows them AI-generated content, August 2, 2026 is your date, and the July 20 guidelines plus the confirmed Code of Practice are your compliance map. The work is mostly inventory, disclosure copy, and marking mechanics: unglamorous, bounded, and much cheaper done before the deadline than after.