-

IRIS likely does not violate Article 5 if it strictly detects physiological states (e.g., eye closure, blink rate). The legal difference is that physiological detection is observable and measurable, whereas emotion inference involves subjective interpretation, which the AI Act restricts in certain contexts.

Likely no by default: detecting physiological drowsiness is generally distinguishable from inferring emotional states, but yes it could fall under Article 5 if IRIS is framed or used to infer emotions (e.g., stress, mood, engagement) in prohibited settings; legally, the key difference is objective, measurable physiological signals versus subjective psychological inference, with the latter triggering stricter AI Act prohibitions.

Yes

IRIS detects physical signs of fatigue, such as eye movement, rather than interpreting emotions. This focuses on observable safety indicators instead of guessing a person’s mental state. Because of this distinction, it is generally seen as physiological monitoring rather than prohibited emotional inference.

IRIS is not likely to fall under the EU AI Act ban if it detects objective physiological drowsiness rather than inferring subjective emotions, with the legal distinction being measurement of physical state versus interpretation of emotional intent.

They are both interlinked and some may argue that emotional state can be passive to a physical state eg lack of concentration and drowsiness due to fatigue

Under the EU AI Act, IRIS is legally permitted because detecting physiological fatigue for safety is distinct from prohibited "emotion recognition" (Article 5), which is used to judge character. However, if the system causes harm or shows demographic bias, owners face strict accountability under the AI Liability Directive, which treats faulty detection as a product defect. To comply, IRIS must maintain Technical Documentation (Articles 11 & 18) for 10 years and complete a DPIA and Fundamental Rights Impact Assessment. Ultimately, standard EULAs and Privacy Policies cannot "shield" owners from liability, as EU law prevents using contracts to waive responsibility for personal injury or discriminatory outcomes caused by AI.

It should be made clear that a physiological state detected in this scenario does not imply an emotional state.

Article 5 of the EU AI Act prohibits certain uses of AI for emotion recognition, particularly in sensitive contexts such as workplaces or educational institutions. A key legal distinction exists between: Detecting a physiological state (e.g. eye closure, yawning frequency, heart rate variability), and Inferring an emotional or psychological state (e.g. stress, anxiety, intent). Drowsiness is more plausibly framed as a physiological and cognitive impairment state, not an emotional one. If IRIS strictly limits itself to observable physical indicators related to alertness, it is less likely to fall under the emotion-inference prohibition. However, risk arises if: The system claims to infer emotions, or Marketing or documentation frames fatigue as an emotional or mental state. Clear technical and legal framing is therefore essential to avoid classification under prohibited practices.

Yes, definitely.

Yes you can test

Shiii idk

IRIS is unlikely to fall under the EU AI Act’s Article 5 prohibition, provided it is clearly framed and implemented as detecting a physiological safety state, not inferring emotions

Well, something in such state shouldn't be driving in the first place, because they pose to be a danger to other road users. I don't think this should fall under the prohibition.

YES. It may be considered a psychological state from intent.

It's hard to give a definitive answer as there would undoubtedly be unique and nuanced situations. My answer would be that it does fall under that prohibition. A legal difference could be visible signs of trauma like injuries. Then again, some physiological states could be a result of an emotional state and vice versa. It would be difficult to distinguish between the two.