What does the AI ​​Act provide, the first law in the world on artificial intelligence approved by the EU

The European Parliament definitively approved theAI Act, the artificial intelligence law: will guarantee the privacy and fundamental rights of European Union citizens and promote innovation. The AI ​​rule has been approved Wednesday 13 March with 523 votes in favor, 46 against and 49 abstentions: the EU thus becomes the first institution to regulate the rules on the use of AI. The text had previously been approved on June 14th. With this approval the European Union – which in recent months has proven to be very attentive to the risks of AI – is the first institution to regulate artificial intelligence systems; however, before entering into force the text will need to be approved again by the Council of the European Union.

The MEP Brando Benifeirapporteur in the European Chamber for the AI ​​Act, declared:

We finally have the world’s first binding law on artificial intelligence, to reduce risks, create opportunities, fight discrimination and bring transparency. (…) The AI ​​Office will now be established to support companies to start complying with the rules before they come into force. We have ensured that human beings and European values ​​are at the very heart of AI development.

The AI ​​Act will come into force later 20 days from its publication in Official Journal of the European Union and will be applied after a further 2 years. For applications classified as “unacceptable risk” it would seem that the bans could start as soon as possible 6 months.

But how does this new law work specifically? The text provides that the applications of artificial intelligence are classified into risk levels: specifically minimal, limited, high and unacceptable risk. One is associated with each category list of obligations on the basis of possible risks for the user or for the context in which that given application is used (think of the healthcare sector or those relating to the stability of a State). Each AI tool classified as “unacceptable risk” will be banned.

Applications associated with an unacceptable risk, and therefore prohibited, are those that involveimage acquisition from surveillance cameras or the internet in order to create databases for the recognition facialas well as the systems of automatic emotion recognition based on facial expressions or sui proxemics workplaces or within the schools.

risk pyramid ai act

For them police it will not be possible to use tools biometric identification in real time in the absence of judicial or administrative authorization, except in cases of extreme necessity such as the search for a missing individual or the prevention of terrorist attacks.

It will also be mandatory for EU countries to establish national supervisory agencies make available to small and medium-sized businesses experimentation spaces within which companies can test the development of systems based on artificial intelligence before placing them on the market.

Finally, the AI ​​Act also establishes that all deepfake – videos in which one person’s biometric data is superimposed on another’s – must be explicitly labeled as such. It will therefore no longer be possible to create altered content without informing those who use such content of the alteration.