Key points in the EU's historic deal on AI governance

Lawmakers will continue to discuss detailed regulations in the coming weeks so that the final process can be completed early next year, expected to apply from 2026.

Until then, companies are encouraged to sign up to the voluntary AI Pact to implement the rules' key obligations.

Below are the main contents of the agreement agreed by the EU.

High risk system

So-called high-risk AI systems – those deemed likely to cause significant harm to health, safety, fundamental rights, the environment, democracy, elections and the rule of law – will have to comply with a range of requirements, such as undergoing tests assessing the impact on fundamental rights and EU market access obligations.

Huenit, an AI camera mounted on a robotic arm at a consumer technology fair held in Berlin on September 1, 2023. Photo: Reuters

Low-risk systems, meanwhile, would be subject to lighter transparency obligations, such as labeling AI-generated content so users can consider using it.

AI in law enforcement

Law enforcement agencies are only allowed to use real-time remote biometric identification systems in public spaces, to identify victims of kidnapping, human trafficking, sexual exploitation and to prevent specific and imminent terrorist threats.

Authorities will also be allowed to use AI technology to track suspects of terrorism, human trafficking, sexual exploitation, murder, kidnapping, rape, armed robbery, participation in a criminal organization and environmental crimes.

General and Platform AI (GPAI) Systems

GPAI and the underlying models will be subject to transparency requirements such as producing technical documentation, complying with EU copyright law, and publishing detailed summaries of the content used to train the algorithms.

Platform models that fall under the category of potentially creating systemic risk and high-impact GPAI will be required to conduct a general model assessment, review and mitigate risks, conduct reverse engineering, notify the European Commission of serious incidents, ensure cybersecurity and report on energy consumption.

Robot "Rmeca" at the Good Global Summit in Geneva, Switzerland. Photo: Reuters

Until EU harmonised standards are published, GPAIs are at systemic risk of relying on codes of practice to comply with regulations.

Banned AI systems

Prohibited conduct and content include: Biometric classification systems that use sensitive characteristics such as political , religious, philosophical beliefs, sexual orientation, and race;

Non-targeted scanning of facial images from the Internet or CCTV footage to create a facial recognition database;

Emotion recognition in the workplace and educational settings;

Social scoring is based on social behavior or personal traits;

AI systems manipulate human behavior to subvert their free will;

AI is used to exploit human weaknesses due to their age, disability, economic or social circumstances.

Sanctions

Depending on the violation and the size of the company involved, fines will start from 7.5 million euros ($8 million) or 1.5% of annual global turnover, rising to 35 million euros or 7% of global turnover.

(According to Reuters)

EU reaches historic deal to regulate artificial intelligence

An agreement between the European Parliament and member states will regulate artificial intelligence (AI), social media and search engines.

EU reaches agreement on Smart Device Protection Act

On November 30, EU lawmakers reached a preliminary agreement on a proposal to protect connected devices, requiring cybersecurity assurance for products with digital elements before they are circulated for use in this market.

EU gets tough on tech giants, threatens to break up violating companies

The European Union (EU) continues to tighten management of large technology companies with strict sanctions, which can fine 10%-20% of global revenue, and even force the splitting of companies that do not comply with regulations.

Source