The European Union is preparing to enact legislation to regulate artificial intelligence (AI) after reaching an agreement with the European Council. This law aims to control AI usage based on different levels of risk. There are three levels: minimal risk, high risk, and unacceptable risk.
Minimal risk includes systems such as product recommendations or spam filters. These systems do not have any control measures, but users may declare their own guidelines for usage.
High-risk systems involve controlling infrastructure, screening individuals for educational institutions or employment, and systems that recognize and record behavior. These systems must have regulatory measures in place to ensure proper usage.
Unacceptable risk refers to systems controlling public behavior, social scoring systems, or systems predicting various actions. The usage of these systems is strictly prohibited.
In addition to controlling usage, this legislation also requires explicit identification of AI, such as deepfake chats or videos, which must clearly indicate that they are artificially generated. The output of AI, including sound, videos, and text, must contain easily verifiable data.
Another addition is the regulation of General Purpose AI, also known as LLM, which includes risk control, adversarial testing, and monitoring of incidents. The regulations for this group may vary across industries but will be enforced under this law.
Violations under this legislation can result in fines of up to 35 million euros or 7% of a company’s total revenue (whichever is higher) for the usage of prohibited AI systems. Non-compliance with other regulations can result in fines of 15 million euros or 3% of total revenue. In the case of misleading notifications, fines may amount to 7.5 million euros or 1.5% of the company’s revenue. Enforcement of specific bans will take place within 6 months, while some regulations may be delayed for up to two years.
TLDR: The European Union is set to pass laws that control the use of artificial intelligence (AI) based on different risk levels. The legislation includes measures to regulate minimal, high, and unacceptable risk AI systems. It also requires clear identification of AI-generated content and sets fines for non-compliance. The regulations will be enforced within specified timelines.
Leave a Comment