The AI Act: Eight Banned Practices Now Illegal in the EU

Starting February 2, 2025, companies and public institutions must cease using AI systems involving prohibited practices, including social scoring and biometric surveillance, as the EU’s strict new regulatory framework takes effect.

Unacceptable AI Practices

The AI Act strictly prohibits eight categories of artificial intelligence deemed dangerous to fundamental rights, safety, and human autonomy. These bans, effective as of February 2, 2025, apply to harmful manipulation, social scoring, certain predictive policing, mass facial recognition databases, and emotion recognition in workplaces and schools.

Organizations, whether private or public, can no longer justify the implementation of a system based solely on its innovation or efficiency. Compliance now requires confirming that the system’s design and intent do not violate these established prohibitions before deployment.

Legal Accountability for Users

The prohibition extends beyond technology providers to include anyone who deploys or uses these systems. If an institution uses tools to profile citizens or a firm monitors employee behavior through banned AI, both the vendor and the user may face significant legal liability.

Manipulation and Social Scoring

AI systems designed to manipulate human behavior through subliminal techniques or misleading mechanisms that cause significant harm are prohibited. This includes exploiting vulnerabilities related to age, disability, or socio-economic status, which poses a specific risk for entities targeting minors, the elderly, or marginalized groups.

Social scoring, or the classification of individuals based on personal behavior or predicted traits in contexts unrelated to original data collection, is strictly banned. This applies heavily to public administration, finance, insurance, and digital platforms.

Predictive Policing and Facial Recognition

Automated systems that predict criminal intent based solely on personality profiling are prohibited. While law enforcement may still use AI to support human judgment backed by objective evidence, the use of AI for purely abstract criminal profiling is forbidden.

Massive, indiscriminate harvesting of facial images from the internet or CCTV to build facial recognition databases is also outlawed. This impacts businesses relying on bulk biometric scraping for marketing, security, or identity verification.

Workplace and Educational Surveillance

Emotion recognition technology is prohibited in schools and workplaces, except for strictly medical or safety reasons. Employers using AI to track employee stress, engagement, or loyalty, and schools using systems to analyze student attention or emotional reactions, face severe legal risks.

Biometric and Public Security Constraints

Biometric categorization based on sensitive traits like political views, religion, or sexual orientation is banned due to the risk of discrimination. Real-time remote biometric identification in public spaces by law enforcement is generally prohibited, with very narrow exceptions requiring prior judicial authorization.

Penalties for Non-Compliance

The European Commission has provided non-binding guidelines to assist with implementation, but enforcement remains under the jurisdiction of the Court of Justice of the European Union. Violating these bans can result in fines of up to 35 million euros or 7 percent of a company’s total annual global turnover.