The EU's AI Act Just Started Enforcing Itself — What It Actually Requires From Chatbot Apps

On 2 August 2026, the European Commission's AI Office and national market surveillance authorities across EU member states activated full enforcement of the AI Act's transparency requirements. The headline obligation is a simple one: if a user is talking to a chatbot or another interactive AI system, the app has to say so, in the interaction itself, not buried in a terms-of-service page. Deepfakes and AI-generated or altered images, video, and audio now have to carry machine-readable marks so the content can be identified as synthetic, and AI-generated content more broadly has to be labelled as such.

The penalties attached are real enough to change a product roadmap. Violations of the transparency rules can draw fines of up to 3% of an AI provider's global annual revenue; violations involving genuinely prohibited AI practices — the more severe tier — can reach €35 million or 7% of global turnover, whichever is larger. For a consumer-facing app built around an AI companion or assistant feature, that is not a rounding-error risk.

€35mor 7% of turnover, whichever is larger

Not every AI obligation landed on schedule. The EU's Digital Omnibus, signed into law as Regulation 2026/1744 on 27 July 2026, pushed back the conformity-assessment deadline for "high-risk" AI systems — the category covering employment, credit scoring, law enforcement, and healthcare uses — to December 2027. The transparency rules that took effect 2 August were never part of that extension: disclosure and labelling are enforceable now, regardless of what category an app's AI features fall into.

Appeye's own catalogue gives a small, honest window into how much of the app economy this actually touches: 260 apps in the catalogue have "chatbot," "AI companion," or "AI assistant" language in their own name or App Store description, and 86 of those already have a computed Appeye score — averaging 78.1 out of 100, with an average App Store rating of 4.59 stars. We also checked whether users were already flagging AI-transparency issues in their own words — phrases like "didn't know it was AI" or "thought it was a real person" — across the real reviews for these apps. Out of 22,873 reviews, that language showed up in exactly 5. That's too small a signal to call a finding, and we're not going to dress it up as one; if anything, it suggests the disclosure gap the AI Act is targeting hasn't yet become a visible complaint pattern in reviews, whether because most apps already disclose reasonably well or because users simply don't articulate the concern that way.

5 / 22,873reviews mentioning AI-disclosure confusion

What the AI Act changes is the downside of getting this wrong going forward — not a retroactive judgment on any specific app in Appeye's catalogue today, and not something Appeye's own scores currently measure at all.

The practical test is no longer 'does this feel honest' — it's 'can a regulator point to the exact line where we said this is AI.'

The practical test for any AI-companion app operating in the EU is no longer "does this feel honest," it's "can a regulator point to the exact line in the product where we said this is AI" — a much more concrete bar, and one enforcement can now actually check.