In April 2021, the European Commission presented the draft European Union Artificial Intelligence Act (AIA) as a first comprehensive horizontal as part of a plan to be a global leader in the development of secure, trustworthy and ethical artificial intelligence (AI). The AIA pursues a proportionate risk-based approach to regulate AI, which distinguishes AI systems that pose (i) an unacceptable risk, (ii) a high risk, or (iii) a low or minimal risk. Those AI systems that create unacceptable risks are those that contravene Union values, for instance by violating fundamental rights, and are therefore prohibited according to Article 5 AIA. The prohibited AI practices covered by this article include four categories of AI systems, namely (i) those that deploy subliminal techniques beyond a person’s consciousness, (ii) exploit vulnerabilities, and (iii) social scoring systems as well as  4) ‘real-time’ remote biometric identification systems.

The present paper aims to present the key legal, psychological and technical issues related to the first category of subliminal AI systems. These include various legal aspects, such as open questions about the relation of the AIA to other legal regimes and the fundamental rights listed in the EU Charter. The psychological aspects mainly refer to the open question about the threshold of subliminal versus supraliminal perception and the role of the senses individually as well as jointly in trying to establish such threshold. The technical aspects refer to both existing AI systems and related technologies as well as future ones that may be introduced in the market in the near future. Finally, the paper aims to briefly examine the question about the scope of the harm requirement laid down in the AIA and the difficulties that may arise in the enforcement of the said prohibition.