The European Union has reached a provisional agreement on amendments to the AI Act, the comprehensive European legislation on artificial intelligence. Member states and European Parliament negotiators decided, among other things, to delay the entry into force of parts of the regulation and to exempt certain sectors. Critics see this as a clear concession to the power of large technology companies.
According to Reuters, the agreements were reached after lengthy negotiations between representatives of EU countries and the European Parliament. The changes still need to be formally approved before they are definitively implemented.
A key part of the compromise is the postponement of rules for so-called high-risk AI systems. These include applications in biometric identification, critical infrastructure, and law enforcement. While these obligations were originally set to take effect this year, implementation has now been pushed back to the end of 2027.
In addition, manufacturers of machines will be given more leeway. Equipment already covered by existing sector-specific regulations will be kept outside the scope of the AI Act. With this, the European Commission is attempting to accommodate companies that warned of double regulation and additional administrative burdens.
The relaxations are part of a broader strategy by Brussels to make digital regulations simpler and more practical. In recent years, European companies have increasingly complained that new legislation hinders innovation and makes it harder to compete with American and Asian players.
New rules against harmful AI content
At the same time, the EU is actually tightening certain parts of the AI rules. For example, there will be a ban on AI applications that generate sexually explicit images of people without consent. The measure follows recent controversy over deepfakes and AI-generated images spread via social media platforms.
It will also become mandatory to label AI-generated content with a visible watermark or identifier. This requirement is set to take effect in December of this year.
According to Dutch Member of the European Parliament Kim van Sparrentak, the new measure against explicit deepfakes is primarily intended to better protect women and minors from the misuse of generative AI technology.
Despite the relaxations, the AI Act remains one of the strictest international regulations on artificial intelligence. The law was originally drafted out of concerns about the impact of AI on privacy, security, the labor market, and cybersecurity.