On Sunday February 2 the first compliance deadline of the EU AI Act came into force. Applications deemed to pose “unacceptable risk” or harm are now banned in the EU.
While the EU Artificial Intelligence Act came technically into force in August 2024, it is being rolled out in stages over 5 years (see timeline at the end) to both give businesses the time to adjust as well as regulators to prepare.
The risk based approach of the Act, which organizes types of AI applications into risk classes from “unacceptable” over “high” to those application only needing to follow transparency requirements, consequently bans any application in the “unacceptable” category. Those prohibited applications — covered in Article 5 of the Act¹ — are:
Companies or services that are found to be using any of the AI applications in this category in the EU will be subject to fines. Like for the GDPR, this applies regardless of where they are headquartered, as long as those services are provided in or target the EU market.
Fines can be up to €35 million or 7% of their annual revenue from the prior fiscal year, whichever is greater.
Some people have criticised that the category “behavioural manipulation” is to ill defined and overall too many loopholes still exist, but only the enforcement in practice will show how the regulators and courts will interpret the scope of those clauses in the AI Act. The focus here will certainly be on “harm”, intrusion of privacy and self determination of the individual.
Another point of criticism has been that the ban on “biometric identification” comes with some exceptions and the term “real time” is vaguely defined.
Some exceptions may be allowed for law enforcement purposes. “Real-time” remote biometric identification systems will be allowed in a limited number of serious cases, while “post” remote biometric identification systems, where identification occurs after a significant delay, will be allowed to prosecute serious crimes and only after court approval.²
However Article 5 also provides for transparency requirements, which will, over time, certainly help to judge if the definition of “real time” is strict enough or contains exploitable loopholes:
7. The Commission shall publish annual reports on the use of real-time remote biometric identification systems in publicly accessible spaces for law enforcement purposes, based on aggregated data in Member States on the basis of the annual reports referred to in paragraph
It is worth nothing that the AI Act, differently from the GDPR, does apply to government institutions and law enforcement and is not limited to business. There will certainly be a lot of struggling around the exceptions for law enforcement in the next years, with law enforcement agencies trying to push the limits of what the AI Act allows them to do.
Despite the vagueness in some aspects, with the deadline having passed, the European Union is drawing a clear line AI companies may not cross in the EU, red lines that currently do not exist either in the US or the UK.
¹ EU AI Act, Article 5 — Prohibited AI Practices
² EU AI Act: first regulation on artificial intelligence — EU parliament