Permitted & Prohibited Uses
🖱️ Interaktiv: Ziehen zum Drehen · Lektion 3
Learning objectives
After this lesson you will be able to:
- Name the AI practices prohibited under Art. 5 EU AI Act
- Recognize examples of permitted but regulated AI use
- Correctly label deepfakes and synthetic content
- Understand exceptions to prohibitions in law enforcement
Prohibited AI practices (Art. 5)
The EU AI Act prohibits the following AI practices — regardless of who deploys them:
1. Social scoring
Rating people by government agencies based on their behavior or personality traits over a certain period, when this leads to unjustified or disproportionate treatment.
2. Manipulative AI systems
Systems that use subliminal techniques or targeted manipulation to influence people's behavior — especially where significant harm may result.
3. Exploiting vulnerabilities
Systems that exploit vulnerabilities of people due to age, disability, or economic hardship in order to manipulate their behavior.
4. Real-time biometrics in public spaces
The real-time capture and analysis of biometric data by law enforcement in public spaces — with very narrow exceptions.
💡 Not everything is banned — but much is regulated
• Only the extreme cases are truly prohibited • Most AI applications are permitted but regulated • High-risk systems are not banned — they face strict requirements • The main risk for companies lies in negligence, not in prohibition
Permitted use with conditions
Emotion recognition
Since August 2026, emotion recognition in workplaces is banned. In other contexts it is possible under strict conditions:
- Informed consent of the affected person
- Documentation of how it works and its accuracy
- Regular reviews for bias
Biometric categorization
Categorizing people by gender, race or ethnic origin is fundamentally prohibited. Exceptions apply only for specific, legally defined purposes.
📝 Schnellprüfung
May a company use AI-based emotion recognition on employees?
Deepfakes and synthetic content
Creating deepfakes (AI-generated image, video or audio content that imitates real people) is not fundamentally prohibited, but is subject to strict labeling obligations:
- Transparency: deepfakes must be clearly labeled as AI-generated
- Exceptions: satire, parody, artistic works (where there is no intent to deceive)
- Prohibited: deepfakes that incite criminal acts or manipulate elections
🏢 Praxis-Szenario: A deepfake in marketing
A marketing department creates an AI-generated video in which the CEO presents product news. The video looks deceptively real. Some customers believe the CEO actually recorded it. How should this be classified?
Law enforcement and national security
The EU AI Act allows narrow exceptions for law enforcement:
- Prosecuting serious crimes such as terrorism or human trafficking
- Preventing immediate danger to life and limb
- Judicial approval required
- Time-limited to 48 hours (real-time biometrics)
💡 Relevance for companies
Most prohibitions concern government agencies or specific use cases. For typical business use (ChatGPT, Copilot, image generation), the transparency and documentation obligations matter more than the prohibitions.
✅ Wichtige Erkenntnisse
Haken setzen, um deinen Lernfortschritt zu markieren:
→ Go deeper: EU AI Act & risk classes — the risk classes at a glance · Transparency obligations — labeling duties for deepfakes
Weiterführende Inhalte
Relevante Rechtsgrundlagen: Art. 5