AI systems with ‘unacceptable risk’ are now banned in the EU

In the European Union until Sunday, block controllers can ban the use of AI systems as “unacceptable risks” or the cause of loss.

The first consent period for February 2 I have actingThe European Parliament is finally the AI ​​regulatory structure of the AI ​​regulatory structure approved last March after the development of the year after year. The law was officially effective August 1; The first one of the consent period is now following.

The specific details have to be set in paragraph 5However, this law is designed to cover in countless use where AI may appear and communicate with people from consumer applications to physical environment.

Under BlockThere are four broad risk levels: (1) minimum risk (eg, email spam filter) will not face any regulatory supervision; (2) Limited risk, which includes customer service chatboats, will oversee a light-tatch regulator; (3) High risk – AI is an example for healthcare recommendations – will face heavy regulatory supervision; And (4) unacceptable risk applications – Focus on this month’s consent requirement – will be completely banned.

Some unacceptable activities contain:

• AI is used for social scoring (eg, creating risk profiles on the basis of a person’s behavior).
• AI which a person’s decisions are neglected or disguised in a manner.
• AI which uses weaknesses such as age, disability or socioeconomic conditions.
• AI which tries to predict the crime collision based on their presence.
• AI that uses biometrics to guess the characteristics of a person as their sexual attraction.
• AI which collects “real time” biometric data in public for the implementation of law.
• AI that tries to guess people’s emotions in action or school.
• AI that scraps the images online or from the protection camera creates facial recognition databases – or expands.

Companies that have been seen using any of the above AI applications in the EU will be subject to fine wherever they are. They can be above the hook for € 35 million (~ 36 million), or their annual income from the previous fiscal year, any higher.

In an interview with TechCrunch, Rob Sumro, the technology chief of the British law firm Slaughter and May, mentioned that the fine would not be kicked for some time.

“Companies are expected to be fully adhered to by February 2, however … the companies need to be aware of the next big deadline are in August,” said Sumro. “By then we will know who the skilled authorities, and the provision of fine and implementation will be effective.”

Initial promise

February 2 deadline is a formality in some ways.

Last September, more than 100 companies signed I agree to AIA voluntary commitment to start implementing the AI ​​Act principles before entering the application. As part of the agreement, the signatories – which included Amazon, Google and OpenAIs were included – committed to identifying AI systems as likely to be classified as high risk under the AI ​​Act.

Some technology giants, especially Meta and Apple, have avoided the deal. French AI Startup MistralThe AI ​​has decided not to sign among one of the strict critics of the Act.

It is not suggested that Apple, Meta, Myster or others who did not agree to this agreement would not fill their obligations – including ban on unacceptable risky systems. Sumoro mentions that the nature of the prohibited use case is given, most companies will not be involved in that practice anyway.

“For companies, the main concern around the EU AI law is that obvious guidelines, criteria and code of conduct will be present on time – and severely, whether they will provide clarity to the agencies,” said Sumoro, “Sumro said. “However, working groups so far… developers meet their deadline about the code of conduct.”

Exemption

There are exceptions to several sanctions in the AI ​​Act.

For example, law enforcement allows law enforcement systems that help to perform any “target search”, to help the abducted victims or a “specific, sufficient, and” to prevent threats. This exemption requires approval from the appropriate management committee, and the law emphasized that law enforcement could not make decisions that “create an adverse legal impact on someone based on the results of these systems.”

This law also creates exceptions to systems that assume emotions in work and schools where there is a “treatment or protection” justice like the therapeutic use of therapeutic use.

European Commission, EU executive branch, Said that it would reveal additional guidelines “In the beginning of 2025” after consulting with stakeholders in November. However, those guidelines have not yet been published.

Sumro says that the other laws of the books are also unclear how to contact the AI ​​Act restrictions and related provisions. The applicable window may not be clear until the end of the year as it comes closer.

“It is important for companies to remember that AI does not exist in control of control,” Sumroo said. “Other legal structures, such as GDPR, NIS 2, and Dora, AI Act, will create potential challenges – especially around the notifications for overlapping events. Understanding how these laws fit together together are as important as the AI ​​law itself. “