Prohibited AI Practices Under EU AI Act Article 5: Analysis and Risk Safeguards

Article 5 of the European Union AI Act defines a categorical ban on specific artificial intelligence practices deemed to pose unacceptable risks to human safety, livelihoods, and fundamental rights. Unlike High-Risk AI systems under Annex III—which are permissible provided they meet strict compliance standards—systems falling under Article 5 are prohibited from development, importation, distribution, and use within the EU.

Comprehensive List of Banned AI Practices

Compliance and engineering teams must verify that active product roadmaps do not incorporate any of the following eight prohibited categories:

1. Subliminal and Manipulative Techniques (Article 5(1)(a)): AI systems deploying cognitive behavioral manipulation beyond a person's consciousness to distort behavior and cause significant harm.
2. Exploitation of Vulnerabilities (Article 5(1)(b)): Systems exploiting specific vulnerabilities of individuals or groups due to age, disability, or social and economic situations to induce harmful actions.
3. Biometric Social Scoring (Article 5(1)(c)): Evaluation or classification of individuals or groups over time based on social behavior or personality traits leading to detrimental or disproportionate treatment.
4. Individual Predictive Policing (Article 5(1)(d)): Risk assessments predicting the likelihood of an individual committing a criminal offense based solely on profiling or personality traits (excluding predictive maintenance of criminal evidence).
5. Untargeted Scraping for Facial Recognition (Article 5(1)(e)): Creation or expansion of facial recognition databases through untargeted scraping of facial images from the internet or CCTV footage.
6. Workplace and Educational Emotion Recognition (Article 5(1)(f)): AI systems inferring emotions of natural persons in workplaces and educational institutions, unless implemented strictly for medical or safety reasons.
7. Biometric Categorization of Sensitive Attributes (Article 5(1)(g)): Systems deducing race, political opinions, trade union membership, religious or philosophical beliefs, or sexual orientation from biometric data.
8. Real-Time Remote Biometric Identification (Article 5(1)(h)): Real-time biometric scanning in publicly accessible spaces for law enforcement, except under narrow, judicially authorized emergency exceptions.

High-Risk Gray Zones and Architectural Guardrails

Enterprise products often inadvertently touch prohibited categories without malicious intent:

• Customer Sentiment vs. Emotion Recognition: Analyzing sentiment from customer support emails or audio transcripts for quality control is generally permitted, provided it does not infer deep psychological or emotional states in employment contexts.
• Computer Vision and User Analytics: Utilizing cameras in physical retail or workplace environments to track movement must ensure facial recognition models do not run untargeted feature extraction or demographic profiling.

Immediate Action Required: Prohibited Practice Audit

Violations of Article 5 carry the most severe enforcement penalties under the EU AI Act: fines of up to €35 million or 7% of total worldwide annual turnover, whichever is higher. Every enterprise deploying computer vision, audio processing, or user behavior analytics must conduct an exhaustive exclusion audit to ensure no deployed software intersects with prohibited techniques.