On 2 February 2025, we saw the first provisions of the EU AI Act become applicable.

Article 5 prohibited AI practices

Article 5 is grabbing the headlines as this outright prohibition bans the most (in the EU's view) dangerous types of AI systems. While some are thinking of futuristic robocops and other sci-fi related examples, the reality is these systems already exist and in the mind of the EU could cause such significant harm to fundamental rights or the safety of an individual that they are no longer permitted. 

The eight prohibitions are:

1. Subliminal, manipulative or deceptive techniques, e.g. imperceptible audio or visual stimuli to influence consumer purchasing decisions without their awareness.
2. Exploitation of vulnerabilities, e.g. a chatbot designed to exploit the emotional state of individuals in debt to sell high-interest loans. 
3. Social scoring, e.g. a mobile app that tracks users' social media interactions and assigns a score that affects their ability to rent a house or get a job.
4. Predictive policing, e.g. predicting potential criminal behaviour based on an individual's demographic data and past activities.
5. Untargeted scraping for facial recognition databases -remember the various Clearview AI fines...e.g. the Garante's €20 million fine, or the more recent Dutch DPA's fine of €30.5 million. 
6. Emotion inference in sensitive settings, i.e. education institutions and the workplace except if for safety reasons, e.g. AI used to monitor employees through wearable devices inferring employee engagement with tasks or in a meeting. 
7. Biometric categorisation to infer characteristics, e.g. analysing customer voice patterns to infer honesty, anger or stress levels (remember Budapest Bank's fine?) 
8. Real-time biometric identification in public places for law enforcement purposes, e.g. real-time facial recognition systems in airports, train stations and out on public street flagging matches with individuals in criminal databases.

For further details see our article here.

With eye-watering fines for those failing to comply coming into effect on 2 August 2025 it has been a busy time for those already using AI systems - auditing and understanding what systems are in place and whether they fall into the prohibited AI systems category - and if so what to do about it! 

If you'd like to know more about the AI audit services we can provide along with our technology partners - helping you to map your current AI usage, identify areas of risk and non-compliance with regulatory requirements, as well as providing recommendations to manage any identified risk and non-compliance - do drop us a line or get in touch with your usual LS contact. We can also provide operational support to implement the recommendations and manage risk going forwards leveraging various technology solutions.

Article 4 AI literacy

It is important to note that the wide-ranging Article 4 obligation to take measures to ensure AI literacy is also applicable.  We are helping a number of clients to train their employees to ensure compliance. We are still waiting for the code of practice to be released (the rumors circulating suggest the end of February but so far this is just hearsay!) and until we have some more substance to the general provision we don't expect to see much in the way of enforcement. If you are interested in learning more about how we can support you to meet this obligation please get in touch with your usual Lewis Silkin contact.