N※N

EU AI Act: A Risk-Based Approach to AI Regulation

  //science-technology.news-articles.net/content/2 .. -act-a-risk-based-approach-to-ai-regulation.html
  Published in Science and Technology on Sunday, April 26th 2026 at 15:46 GMT by BBC
      Locales: LEBANON, ISRAEL

The Risk-Based Hierarchy

At the core of the AI Act is a tiered approach to risk management. Rather than applying a blanket set of rules to all software, the legislation divides AI systems into four distinct categories:

• Unacceptable Risk: These systems are deemed a clear threat to the safety, livelihoods, and rights of people and are consequently banned. This includes AI that engages in cognitive behavioral manipulation or exploits vulnerabilities in specific groups.
• High Risk: Systems that have a significant impact on a person's life chances or safety--such as AI used in critical infrastructure, education, healthcare, or law enforcement--are permitted but subject to stringent obligations. These include mandatory risk assessments, high-quality data sets to minimize bias, and human oversight.
• Limited Risk: This category applies to AI with specific transparency obligations. For instance, users must be made aware when they are interacting with a chatbot or when content is AI-generated (deepfakes).
• Minimal Risk: The vast majority of AI applications currently in use, such as AI-enabled video games or spam filters, fall into this category and remain largely unregulated.

Prohibited Practices and "Red Lines"

One of the most contentious and significant aspects of the AI Act is the establishment of "red lines"--technologies that are strictly prohibited within the EU. The legislation specifically targets practices that infringe upon fundamental human rights. Prohibited uses include:

• Social Scoring: The use of AI by governments to rank citizens based on their social behavior or personal characteristics.
• Biometric Categorization: Systems that categorize individuals based on sensitive characteristics such as political, religious, philosophical beliefs, sexual orientation, or race.
• Emotion Recognition: The deployment of AI to detect emotions in workplaces or educational institutions, viewed as an invasion of privacy and psychological autonomy.
• Predictive Policing: AI systems that attempt to predict the likelihood of an individual committing a crime based on profiling.

General Purpose AI and Foundation Models

The act also addresses the rise of General Purpose AI (GPAI) and large-scale foundation models, such as those powering ChatGPT and Gemini. Because these models can be integrated into a vast array of different applications, they are subject to specific transparency requirements. Developers of these models must provide technical documentation and comply with EU copyright law, ensuring that the data used to train these models is documented and legally sourced.

For "systemic? models--those with high computing power that could pose systemic risks--additional obligations apply, including the requirement to perform model evaluations and report serious incidents to the European AI Office.

Global Implications and Enforcement

While the legislation is European in origin, its impact is global. Much like the General Data Protection Regulation (GDPR), the AI Act is expected to create a "Brussels Effect," where international companies adapt their global products to meet EU standards to maintain access to the European market.

Non-compliance carries heavy penalties. Fines for violating the most critical rules can reach up to EUR35 million or 7% of a company's total global annual turnover, whichever is higher. This ensures that the cost of negligence is prohibitively expensive for even the largest tech conglomerates.

Summary of Key Provisions

• Risk-Based Approach: Regulations are scaled based on the potential harm to humans (Unacceptable, High, Limited, Minimal).
• Human Rights Protection: Bans on social scoring and emotion recognition in schools and workplaces.
• Transparency Mandates: Clear labeling for AI-generated content and chatbots.
• GPAI Requirements: Foundation model developers must adhere to copyright laws and technical documentation standards.
• Stiff Penalties: Fines up to 7% of global turnover for the most severe infractions.
• Governance: Establishment of the European AI Office to monitor and enforce the rules.