EU’s EU systems with ‘unacceptable risk’ are now forbidden

On Sunday, the regulators of the bloc in the European Union can prohibit the use of AI systems they consider “unacceptable risk” or harm.

The first match for February 2 is the date of I have an actA comprehensive AI regulatory database approved by the European Parliament in recent March after development in recent years. The law officially entered into force 1 AugustOpen Now the following is the first of the compatibility period.

Features are built in Article 5However, ACT is designed to cover countless use, which can be seen with persons from the EU physical environment and can be seen with individuals and interacts.

Under Block approachThere are four extensive risk levels: (1) minimal risk (eg, e-mail spam filters) will face any regulation control; (2) The limited risk included in Customer Service ChatBots will have a light touch regulatory control; (3) High risk – AI is an example for healthcare recommendations – to face heavy regulatory control; and (4) Unacceptable risk applications – focusing on compliance requirements of this month will be completely banned.

Some activities that are not accepted include:

• AI is used for social arm (eg risk profiles based on a person’s behavior).
• AI that manipulates a person’s decisions subliminal or deceitful.
• AI, which exploits weakness such as age, disability or socio-economic status.
• AI is trying to commit crimes based on their appearance.
• AI that uses biometrics to express a person’s characteristics such as sexual orientation.
• AI, which collects “real-time” biometric information in public places for law enforcement agencies.
• AI is trying to imagine the feelings of people in business or school.
• Bringing pictures from online or security cameras – or expanding AI – or expands.

The companies used by any of the EU applications above the EU will be fined, no matter where the headquarters are. They can be in a fork with 35 million euros (up to ~ 36 million euros) or 7% of its annual income from the previous fiscal year.

Fines for some time will start, Rob Sumroy, the head of the British law firm, in an interview with TechCrunch.

“Organizations are expected to be fully compatible until February 2, but … The next major deadline that companies need to be aware will be in August,” he said. “At that time, we will know who the authorities are and who will enter into force of fines and executive provisions.”

Initial Perdas

The date of February 2 is in a certain formality.

Signed more than 100 companies last September I ai pactA volunteer pledge to start applying the principles of movement before accessing the application of the EU action. As part of the Pact, Amazon, Google and Openai signed those who are committed to the identification of AI systems, which are likely to classify as a high risk under the AI ​​Act.

Some technological giants jumped into the pact, especially meta and apples. French ai start MistralOne of the harshest critics of the EU movement preferred not to sign.

This should not offer that Apple, Meta, Mistral or Pacts will not meet the obligations of others who disagree or ban on recipient risky systems. Sumroy will not deal with these practices, taking into account the nature of the prohibited use.

“Although the EU AI ACT is a basic concern around the ACT, the timely and numerous rules of the rules of knowledge, standards and behavior, this is due to inconsistency,” he said. “However, working groups respond to the codection of behavior for developers so far … to date.”

Possible freedoms

There are exceptions for a few of the EU ACT bans.

For example, the ACT allows law enforcement to use certain systems that collect biometrics in public places, if these systems help prevent a hijacking victim or a “specific, fundamental and unavoidable” threat. This freedom requires permission from the relevant management body, and the ACT notes that the law enforcement will not only make a decision of the “negative impact” based on the performances of these systems.

The act also wakes up for systems that are based on the “medical or security” based on the “medical or security” justification, such as the systems designed for therapeutic use.

European Commission, EU Executive Power, said to broadcast additional guidelines After consulting stakeholders in November, “in the early 2025”. However, these instructions were not published yet.

Sumroy said that the books are clear, how it is clear that other laws may link to the AI ​​Act and associated with related provisions. As the executive window approaches, clarity cannot come by the end of the year.

“It is important to remember that the AI ​​regulation for organizations isolated,” he said. “Other legal frameworks such as GDPR, NIS2 and Dora, will interact with the AI ​​Act – especially the notice requirements of the incident.