EU AI Act: GPAI and transparency obligations enforceable since August 2, 2026
As of August 2, 2026, another stage of the EU AI Act has gone live: the obligations for general-purpose AI models (GPAI) and the transparency duties under Articles 5 and 50 are now applicable and enforceable. What was previously an announced roadmap becomes law with a sanction framework. At the same time, the EU pushed the most demanding obligations for high-risk systems to December 2027. For agencies and brands, this means the debate about AI labeling is no longer a future question — it is a matter of ongoing compliance.
What has applied since August 2, 2026
- GPAI obligations: providers of general-purpose models must meet transparency, documentation, and copyright requirements.
- Transparency duties (Art. 50): labeling of synthetic media — AI-generated or manipulated images, audio, and video (deepfakes), plus disclosure for AI chatbots.
- Prohibited practices (Art. 5): the bans in force since February 2025 remain in effect and are being enforced.
- High-risk deferred: the obligations for high-risk systems only kick in from December 2027.
- Enforcement: an announcement becomes binding law — with a fines framework.
Where things stood
The EU AI Act took effect in stages. The bans on certain practices (Art. 5) have applied since February 2025. For the transparency duties under Article 50, August 2, 2026 was set as the application date — until then they were adopted but not enforceable. During that phase, one question circulated in marketing teams above all: must AI-generated text be labeled? We covered the answer in an earlier news piece — for classic marketing and SEO text, the disclosure duty under Art. 50(4) does not apply.
What is now in force
1. The duties are enforceable, not merely adopted. The difference matters in practice: from now on, violations of the transparency and GPAI rules can be penalized. Anyone publishing synthetic images, audio, or video must label them in a machine-readable way and make them recognizable to users.
2. Synthetic media and chatbots are the focus. Article 50 targets deepfakes and AI-generated media, plus disclosure when people interact with an AI system. For agencies this directly affects AI images in campaigns, generated voice-overs, video content, and chatbot interfaces on client sites.
3. The text exemption remains — with limits. The exemption for marketing and SEO text described in our earlier news still holds. But it is narrow: for content on matters of public interest, a disclosure duty may apply. Pure product descriptions and ad copy stay exempt; journalistic-looking content on socially relevant topics does not automatically.
4. High-risk obligations are deferred. The most demanding requirements — for high-risk applications — were pushed to December 2027. That buys companies time, but changes nothing about the transparency and GPAI duties in force now.
Reading
The real shift is not in the wording of the articles — that was settled long ago — but in the status change from “announced” to “enforceable”. That is the point at which many organizations first react. The impression from practice: in many marketing departments, labeling synthetic media has so far been treated as a side issue. With applicability, it becomes a process question.
The two-part timing is notable: the comparatively implementable transparency duties apply now, the complex high-risk requirements only in 2027. That is a pragmatic design — it defers the heavy load and first makes enforceable what can be done with manageable effort. For day-to-day content production, this means the pressure to act today is on labeling and disclosure, not on elaborate risk assessments.
Important for context: this is not legal advice. The concrete set of duties depends on the individual case — the points listed are the framework, not an assessment of your specific situation.
What you can do now
If you publish AI-generated media: introduce labeling for AI images, audio, and video — visible to users and, where possible, machine-readable. This is the point with the highest immediate pressure to act.
If you run chatbots: make sure users recognize they are talking to an AI system. The disclosure belongs at the start of the interaction.
If you own content processes: clarify per content type whether it falls under the exemption. Pure ad copy is exempt — content on matters of public interest you should check separately.