EU AI Act finally comes into first effect

Although the timing is symbolic, this initial phase only applied to certain prohibited uses of AI.

The implementation of the legislation, considered the most ambitious AI regulation in the world to date, was set to be gradual and remained a subject of debate, with some aspects facing challenges. These concerns were raised at the summit, where the Élysée Palace hosted 100 heads of state, including representatives from the United States and China, alongside civil society players and AI industry leaders such as Elon Musk, Sam Altman (OpenAI), and Sundar Pichai (Google).

Key provisions in effect

From 2nd February, specific AI applications deemed unacceptable under the AI Act were officially banned. These included:

• Social rating software, whether public or private, similar to China’s system.
• Individual predictive policing AI designed to estimate a person's likelihood of committing offences.
• Emotion recognition tools used in workplaces or educational institutions to analyse employee or student behaviour.
• AI systems exploiting people's vulnerabilities, as well as those using manipulative or subliminal techniques.
• The real-time identification of individuals via facial recognition in public spaces.
• Biometric categorisation of individuals to infer characteristics such as race, political or religious beliefs, sexual orientation, or trade union membership.

Additionally, the European Commission has committed to releasing a repository of AI literacy practices based on input from providers and deployers. Article 4 of the Act required companies using AI to ensure their employees had sufficient technical knowledge of the AI system, an understanding of its use, and awareness of the potential impact on those affected.

A gradual implementation

While these initial prohibitions were significant, the broader and more impactful provisions of the AI Act were scheduled to take effect in six months. On 2nd August, regulatory authorities in each EU member state are to be appointed, gaining the power to issue fines and enforce compliance. The list of banned AI practices is also subject to an annual review by the European Commission, with further obligations for providers of general-purpose AI models taking effect at the same time.

Beyond these prohibitions and literacy requirements, the regulation also provided a formal definition of AI systems. It preceded the much-anticipated Code of Practice for General-Purpose AI Models, which had been undergoing revisions, with the final version expected in April.

Companies found violating these regulations within the EU risked fines of up to 7% of their annual revenue or €35 million, whichever was greater. However, the law allowed for certain exceptions in law enforcement-related activities, such as real-time facial recognition in public spaces or AI-assisted suspect assessments in criminal investigations.