Your journaling app probably isn't "emotion recognition" under the EU AI Act

Everyone building anything emotional with AI reaches for the emotion recognition article. The sentence that is more likely to apply to you is in Article 50(1), and it has applied since 2 August.

When I went looking at what the EU AI Act asks of an app like Murror, I assumed the provision about emotions was the one aimed at us. It isn't, quite, and the one that is aimed at us is simpler and already live.

Start with the one people cite. Article 50(3) requires deployers of an "emotion recognition system" to inform the people exposed to it. The definition in Article 3(39) is an AI system for identifying or inferring emotions or intentions of natural persons, and the wording goes on to say that the inference is based on their biometric data. Face, voice, gait and the like. An app that reads text a person chose to type is, on my reading, not obviously in that box. I'm not a lawyer and this is my reading of the text, not advice, but it is worth reading the definition yourself before you either panic or relax.

Now the one that likely does apply. Article 50(1) says providers must ensure that AI systems intended to interact directly with people are designed so that the people concerned are informed they are interacting with an AI system, unless that is obvious from the circumstances. If your product has a conversational AI in it, that is the sentence to read. The information has to be given no later than the first interaction.

The timing is the part I got wrong in my head. The Digital Omnibus, Regulation (EU) 2026/1744, delayed the high-risk rules from 2 August 2026 to 2 December 2027, according to the law-firm summaries I read (Orrick and Jones Walker). It did not delay most of Article 50. The only grace period those summaries describe is for the machine-readable marking of synthetic content: providers of systems already on the market before 2 August have until 2 December 2026. I could not get the Official Journal text to load for the omnibus itself, so check the dates against EUR-Lex before you rely on any of this.

What I'd actually do this week, whatever you build:

First, find every place a person could reasonably think they are talking to a human, or to a person-like voice, and make sure the first thing they see says otherwise in plain words. Not a line buried in the terms. The point of the rule is that the person knows in the moment.

Second, separate two questions that get blurred. Does your product infer emotions from biometric data? Probably no for a text app, and yes for anything touching voice tone or facial expression. Does it talk back like a person? That is the one that matters for most of us.

Third, write down your answer to each, with the date. If a regulator or an app reviewer ever asks, a dated note beats a memory.

For us, the honest version is that Murror is clearly an AI from the first screen, and I'd rather over-label than have anyone in a vulnerable moment wonder who is on the other end. That felt like good product sense before it was a legal question, and it's nice when the two agree.

Which of your products would you have put under the emotion rule, and would you have gotten Article 50(1) right?