A timeline for the AI regulation
A timeline for the AI regulation
The European AI Act will take effect on August 1. Six months later, the prohibitions are in force.
July 15, 2024
On August 1, 2024, the European AI Regulation (AI Act) will come into force. From that moment on, there will be a transition period in which organizations will have time to prepare for the obligations in the law. For most obligations, that period lasts between half and three years. Good to know that the obligations will apply retroactively in 2030 to government organizations that already work with AI systems. So there is work to be done.
Timeline
In the coming period, Digitaleoverheid.nl will publish a series of articles about the various risk categories in the AI regulation. The first in the series includes a handy timeline about the AI regulation.
- Mid-July 2024 — Publication of the AI regulation in the European Official Journal
- August 1, 2024 — Entry into force of the AI regulation
- February 1, 2025 — Prohibitions in effect
- August 1, 2025 — General purpose AI model requirements take effect
- August 1, 2026- Most articles including high-risk AI obligations take effect
- August 1, 2027 — Obligations for high-risk AI systems in products in effect
- August 1, 2030- Obligations for AI systems used by government organizations that were already in use before entry into force come into force.
First, the banned AI systems
Within six months, the prohibitions will take effect. That's about prohibited artificial intelligence systems, which the European Union has decided is not in line with European norms and values. Think of social scoring, the automated scoring process to assess social behavior, as is the case in China. AI systems that are intended to manipulate human behavior and AI to predict crime will also be banned in the EU from 1 February next year. These may not be the direct AI systems that municipalities are currently working with. However, it is wise to check whether you are using AI systems that fall into this category. In that case, they must be phased out in time. As a supervisor, the Data Protection Authority monitors whether prohibited AI is being used in the Netherlands.
Next points on the timeline
With 'General purpose AI models', the next dot on the timeline, refers to generative AI: applications such as ChatGPT, which can generate new text, images, code, video, or sound based on prompts, people's orders. The developers of such models must work to comply with the obligations of the AI regulation, including in the field of transparency. The next inventory that governments must make is that of High-risk AI systems that are used within the organization. This will be subject to a number of obligations from 1 August 2026, including a risk management system and technical documentation. Datasets for these high-risk systems must meet various quality requirements. New high-risk AI systems receive a label before they are allowed to enter the market. The Data Protection Authority supervises existing products that process high-risk AI, for which CE marking is not yet mandatory.
Doubts and questions
In case of doubt whether an AI system falls under the high-risk systems, the supervisor can provide an answer. All Member States must have organised supervision by 1 August 2025. Last month, a consulting to the Cabinet about the supervision of AI in the Netherlands, drawn up by the National Inspectorate of Digital Infrastructure and the Data Protection Authority, in cooperation with many other supervisors. In the Netherlands, supervision is as much as possible in line with the existing tasks of market supervisors. So there will not be one AI supervisor for all sectors. 'How do I know how much impact our algorithm has?'(and variations on this question) are constantly being heard by the people responsible for the national algorithm register. The Ministry of Internal Affairs and Kingdom Relations has created an email address for questions about the AI Act: ai-verordening@minbzk.nl.
Share this article
Marjolein van Trigt
Editor
Marjolein van Trigt is digitization editor at Internal Administration
.avif)