EU’s AI Act Faces Criticism and Implementation Timeline

Starting in August, the European Union will implement the AI Act. Although the provisions will be integrated gradually, experts have already criticized the legislation as underdeveloped.

The EU AI Act

On March 13, 2024, the EU Parliament approved a set of fundamental rules governing AI activities. European Parliament President Roberta Metsola hailed the law as groundbreaking, asserting it would foster innovation while safeguarding fundamental rights.

On July 12, 2024, the document was published in the Official Journal of the EU. It will take effect on August 1, 2024.

The main provisions will be fully enforced by August 2, 2026. Until then, they will be gradually integrated at specific intervals:

• February 2, 2025 — Ban on AI systems posing unacceptable risks, such as tools for manipulating or deceiving people, or classifying them by social rating;
• August 2, 2025 — Provisions for regulating general-purpose AI systems;
• August 2, 2026 — Restrictive rules for “high-risk” AI systems, including those involving biometric data, critical infrastructure, and education.

The document categorizes AI systems by risk levels:

• Low: Models like spam filters or AI-supported video games will not be regulated;
• Restrictive: Chatbots and other text and image generation systems will face certain transparency requirements;
• High: Systems used by law enforcement and government agencies or those performing biometric identification and emotion recognition;
• Unacceptable risk: Systems that can deceive or manipulate people, or assess them based on social behavior or personal traits, are prohibited.

Criticism of the AI Act

Critics have pointed out that the rushed attempts to regulate foundational AI models will limit the technology’s use rather than mitigate risks associated with neural networks, reports FT.

Andreas Cleve, CEO of the Danish healthcare startup Corti, noted that compliance costs could reach six figures for a firm with 50 employees. According to the expert, this is an additional “tax” for small businesses.

“I’m concerned about legislation that becomes too complex for a small company to afford. It’s a challenge to raise funds, and now there’s this tax to pay. You also need to spend time understanding it,” commented Cleve.

Kai Zenner, who participated in drafting the rules, stated that the law ended up “quite vague” due to time constraints. Regulators couldn’t agree on some issues, leaving them unresolved, he noted.

Cecilia Bonefeld-Dahl, CEO of DigitalEurope, warned that this approach has led to poorly thought-out rules that will hinder Europe’s attempts to compete with the US in AI.

“Additional compliance costs further reduce our performance. We’ll be hiring lawyers while the rest of the world hires programmers,” she said.

In October 2023, Ukraine’s Ministry of Digital Transformation introduced a roadmap for AI regulation to help companies prepare for the adoption of the corresponding EU act.

Previously, Coinbase CEO Brian Armstrong spoke out against AI regulation and called for the industry’s decentralization.