AI Act: European Regulation on Artificial Intelligence is almost official

European Parliament and Council reached an agreement on the Artificial Intelligence Act. After a long institutional process, started by European Commission’s proposal in 2021, in the last 36 hours of negotiations, the EU defined an innovative framework aimed at guaranteeing artificial intelligence’s use safety, respecting human rights and democracy principles.

Briefly, what does the text say?

The agreement lists a series of instruxtions aimed at preserving individual rights, privacy and preventing possible abuses due to AI systems use in sensitive fields like workplaces, education institutions and public services. The following are prohibited:

• Use of biometric systems that categorize based on sensitive information like political, religious, philosophical belief, sexual orientation, race;
• Not-targeted Facial Imagines scraping, i.e. acquisition of facial expression from internet or video surveillance systems to create face detection database;
• Emotion detection on workplaces and education institutions;
• Social ranking based on social behavior or personal characteristics;
• AI use to manipulate human behavior to bypass free will;
• AI use to exploit people’s vulnerabilities due to age, disability or social or economic situation.

Exceptions are foreseen for law enforcement bodies. Safeguarded are biometric identification systems (RBI) use in public spaces, limited to jurisdictional authorization and for a short, limited list of crimes. “Post remote” RBI would be possible for targeted search of individuals sentenced or suspected of severe crimes, while “real time” RBI would be limited to strict conditions, like targeted search of victims, terror threats prevention and localization of specific suspected criminals.

Artificial intelligence systems considered high risk, due to their possible impact on health, safety, human rights, climate, democracy and rule or law must follow clear duties, like impact evaluation on human rights and citizens’ possibility to complaint and receive explanations on high-risk decisions.

Specific guardrails are established for general artificial intelligence systems (GPAI). Transparency, technical documentation, and detailed synthesis of materials used for training. High-impact, systemic risk GPAI models must follow stricter duties, like model evaluation, systemic risks management and detailed reports.

To support innovation, small and medium enterprises developing AI would be protected from giants’ overpressure. To reach this goal, national authorities promote systems to develop and test artificial intelligence innovations, before they enter market.

Sanctions will be taken in case of norms breach: fines from 35 billion euros or 7% of global revenue at 7.5 billion, or 1.5% of revenue, according to the infraction and enterprise’s dimensions.

What to expect?

In the next phases, AI Act’s agreed text will pass through the Parliament’s formal adoption, through the Home Market and Civil Liberties Commissions’ vote, then the Council’s approval, to become bounding for all EU Members.

When the Regulation will come into force, a new era awaits artificial intelligence, in Europe ad globally. AI Act could be the basis for other global legislations, like the EU’ General Data Protection Regulation (GDPR) set a standard for other global privacy laws.

Mondo Internazionale APS – All Rights Reserved®