One of the longest trilogues in memorywhich should culminate in the approval of the AI Act. The twitter profile @MEPAssistantknown in the European bubble for his satirical posts that reveal what happens in the buildings of Brussels, explains that so far the record belongs to the rules on the “Emission Trading System”so the interinstitutional dialogues lasted 30 hours (albeit with a break).
Records aside, this exhausting negotiation is due to a few conflicting factors: on the one hand, the Spanish presidency wants to finalize the negotiations, on the other, the imminent European elections do not leave too much time after the Christmas break and the fewer pending files there are, the better. Many, especially in civil society, are wondering whether it makes sense to let sleep deprivation be the push that leads to the approval of such an important law for the future of businesses and citizens, closing a path started by the European institutions over five years ago .
In fact, after 22 hours, the official announcement arrived after lunch from Commissioner Breton to break ranks to meet again today, Friday 8th, at 09:00.
Strategikon Security and/or democracy? The structural weaknesses in Italian cybersecurity regulations by Andrea Monti 03 November 2023
AI and prohibited practices
While waiting to have time to study the final text carefully, once approved and published, we report one of the issues on which the co-legislators have been struggling for weeks to find a compromise that does not displease everyone too much: that of prohibited practices.
L’AI Act presenta 4 risk levels:
prohibited practices, high risk, limited risk and minimal risk.
In prohibited practicesalready in the initial version, there were the AI systems using “subliminal techniques” harmful manipulative; That exploit specific vulnerable groups (physical or mental disability); those used by public authoritiesor on their behalf, for social scoring purposes; eh remote biometric identification systems “in real time” in spaces accessible to the public for law enforcement, except in a limited number of cases.
Interview The first Italian bill on AI is ready. Ascani: “He will protect us from deepfakes” by Pier Luigi Pisa 14 November 2023
Parliament, in its official position, called for bans on biometric recognition in real time, without any exceptions. From what Euractiv reported, it seems that “the bans confirmed so far concern manipulation techniques, systems that exploit vulnerabilities, social scoring and indiscriminate scraping of facial images.” The latter should be the result of a recent compromise between Parliament and Council. Self passed to the final vote, it would mean that companies that develop facial recognition systems could no longer use the photos that users upload to social media without asking for their consent.
Parliament received, in the capacity of the two co-rapporteurs Brando Benifei (S&D) and Dragos Tudorache (Renew Europe), a package of Council proposals specifically on prohibited practices.
The Council’s proposal, however, would open the way to many exceptions on very questionable practices, with risks to fundamental rights. Second Daniel Leufer Of Access Now, who had access to the document, law enforcement would retain the right to profile people based on race, political and religious preferences, and union membership whether these characteristics are linked to a suspect or crime. It is true that, with these methods, it would be very easy to end up in a society where discrimination is carried out in an automated way.
The story France and its billionaires want to take back AI from our correspondent Pier Luigi Pisa 17 November 2023
Also on the ban that Parliament had requested on emotion recognition the Council is ready to accept it only in workplaces and schools, but not if used by the police or at borders. If Parliament, e Benifei in particular, they had started with a clearer position, the Council, when it comes to security, is not ready to give in points. On point it is France to lead this consortiumthanks to the fact that it is the country that has been most affected by terrorism in Europe.
Also the ban on predictive policing (remember the movie Minority Report?) requested by Parliament it would be greatly reduced by the Council which would like it in cases of existing suspicions of criminal activity.
Regarding law enforcement use of the real-time biometric recognition, Parliament was ready to allow its use for the most serious crimes such as terrorism, but even in this case the Council would like to have freer hands. In the Commission’s initial proposal, certain crimes were punishable by 3 years in prison (which, as one can imagine, may not have been particularly serious). On this point, research and experience tell us that this technology discriminates more against people of color, especially if they are women, increasing the cases of false positives for them and, therefore, they would be stopped by the police more often.
Digital rights Why is the Italian government on the side of those who don’t want to regulate AI? by Diletta Huyskes 24 November 2023
It should be remembered that in Italy, when the police used this technology in the past, they did not consult the Privacy Guarantor, as they should have, and as a result the Guarantor Authority blocked any operation in this sense.
Speaking of the Guarantor, many privacy professionals would see the Guarantor Authority as the ideal candidate to also regulate AI in Italy, given the long experience in the data sector and new technologies, but at the moment it would seem increasingly difficult. The Parliament had in fact asked for the Supervisory Authority to be independent, as are the Guarantor, AGCOM, AGCM, but it seems that this possibility may fade away. And that would be a shame, because an Authority that depends on the governmentwould be subject to changes in management with every political change, not guaranteeing the continuity that companies would also need.
There are still many issues to be resolved and all of them are very important, however the risk of not finding a compromise now, and waiting until January, with a new presidency, is that everything could be called into question and the timeframe considerably extended with the possibility of having a even worse text in the future.