AI Features That Are Banned Under EU Law: What You Need to Know

Artificial intelligence (AI) technologies are transforming industries and daily life — but they also raise serious ethical, legal, and privacy concerns. To address these, the European Union (EU) has enacted the landmark Artificial Intelligence Act (AI Act), a comprehensive regulatory framework governing the development, deployment, and use of AI across its Member States.

At the heart of the AI Act are strict prohibitions on certain AI features and practices that pose unacceptable risks to individuals’ rights, safety, and freedoms. These bans — now legally binding — set the EU apart globally in its approach to AI governance and have serious implications for companies and developers operating within or targeting the European market.

This article unpacks the AI features banned under EU law, why they matter, and what they mean for businesses and users.

What Is the EU AI Act and Why It Matters

The EU Artificial Intelligence Act is the first comprehensive AI regulation in the world, representing a paradigm shift in how governments govern emergent technologies. Rather than a one‑size‑fits‑all approach, it categorises AI systems by risk levels — minimal, limited, high‑risk, and unacceptable — with a tailored legal regime for each.

Unacceptable risk AI systems are prohibited entirely. These are applications that the EU deems too dangerous to be placed on the market, put into service, or used in society due to fundamental rights violations, safety risks, or other harms.

AI Features and Practices Banned Under EU Law

Below is a comprehensive breakdown of AI features and practices prohibited by the EU AI Act (effective 2 February 2025):

Prohibited AI Feature / Practice	Why It’s Banned	Example or Insight
1. Subliminal & Manipulative Techniques	Distorts user behaviour without informed consent; undermines autonomy	AI that covertly influences purchasing decisions through hidden prompts or nudges
2. Exploitation of Vulnerable Groups	Targets age, disability, or socio‑economic vulnerabilities	Toys that use voice AI to encourage risky behaviour in children
3. Social Scoring	Creates unfair societal stratification and discrimination	Systems that rate individuals’ behaviour to determine access to services or opportunities
4. Predictive Policing based on Profiling	Risks discrimination and rights violations in law enforcement	Predicting crimes based solely on demographic or appearance data
5. Untargeted Facial Image Scraping	Enables mass surveillance and privacy invasion	Scraping public CCTV or web images to create biometric databases
6. Emotion Recognition in Sensitive Spaces	Infers emotions without consent, risking discrimination	AI monitoring employees’ emotions via webcams at work
7. Biometric Categorisation of Sensitive Attributes	Infers race, sexual orientation, religion, etc., from biometric data	Using facial analysis to guess political beliefs
8. Real‑Time Remote Biometric Identification (Public)	Mass surveillance without due process	Live facial recognition in public spaces without strict exceptions

Each of these is prohibited because it violates core EU values, including privacy, equality, dignity, and freedom.

Real‑World Examples and Case Studies

1. AI That Manipulates Human Behaviour

Imagine an online shopping site that uses AI to determine your emotional state and then displays content or prices tailored to push you into buying more. Under the AI Act, this AI would be banned because it uses manipulative techniques beyond conscious awareness, impairing informed decisions.

Case in point: European regulators have explicitly called out AI systems that embed dark patterns to influence purchases — such as deceptive nudge techniques — as unacceptable.

2. Biometric Emotion Detection in the Workplace

Some companies have experimented with emotion‑detecting AI to assess employee engagement or focus during meetings. While such systems may seem innovative, in the EU they fall foul of the prohibition on emotion recognition — especially in workplaces and schools, unless justified for legitimate medical or safety purposes.

3. Social Scoring Systems

Inspired in part by debates around China’s social credit systems, the EU’s ban on social scoring prevents AI from creating opaque behavioural scores that influence access to rights, services, or opportunities. Even private firms risk violating the AI Act if they deploy opaque scoring mechanisms that penalise certain groups.

Why These Bans Are Crucial for Privacy and Trust

The AI Act reflects a privacy‑centric philosophy closely aligned with the EU’s existing data protection framework, especially the GDPR. While the GDPR focuses on data processing and individual rights, the AI Act targets how AI systems can shape behaviour, predict characteristics, and influence decisions — adding an extra layer of protection.

Key reasons for the bans include:

• Preventing discrimination and bias
• Protecting vulnerable populations
• Safeguarding personal autonomy
• Limiting unchecked surveillance and profiling
• Ensuring transparent AI ecosystems

Taken together, these prohibitions set a high trust standard for AI development and use in Europe — and increasingly, globally.

Compliance Implications for Businesses

For companies developing or deploying AI technologies in the EU:

1. Know Your Risk Tier

Not all AI is banned. Systems with minimal or limited risk — such as chatbots or spam filters — remain permissible with compliance requirements. High‑risk systems require rigorous checks, documentation, human oversight, and market surveillance adherence.

2. Conduct an AI Risk Audit

Understanding where your systems fall — prohibited, high‑risk, limited, or minimal — is essential. An AI risk audit can help organisations align with EU law and avoid penalties.

3. Design With Privacy by Default

Implement privacy‑preserving features and transparency disclosures early in the development lifecycle. This approach not only mitigates legal risk but fosters user trust.

4. Prepare for Enforcement and Penalties

Non‑compliance with banned AI practices can lead to substantial fines, including up to 7% of global revenue or €35 million, whichever is higher. European authorities have already begun enforcement efforts.

Frequently Asked Questions (FAQ)

Q1: Are all AI voice assistants banned in the EU?
No — voice assistants themselves are not banned. However, if they use subliminal or manipulative techniques that impair decision‑making, they could fall under prohibited categories.

Q2: Can AI be used for facial recognition at all?
Real‑time remote biometric identification in public spaces is prohibited, except under narrow, court‑approved exceptions (e.g., locating missing persons). Less intrusive forms may be permissible with strict safeguards.

Q3: Does the ban apply to AI developed outside the EU?
Yes. If an AI system is marketed or used within the EU, it must comply with the AI Act — regardless of where it was developed.

Q4: What about emotion recognition in health contexts?
Emotion inference may be allowed if it is essential for medical or safety reasons and meets strict compliance criteria.

The European Union’s AI Act marks a milestone in ethical, legal, and human‑centred AI governance. By banning AI features and practices that manipulate behaviour, exploit vulnerabilities, or undermine fundamental rights, the EU is setting a global standard for responsible innovation.

For developers, businesses, and policymakers, understanding these prohibited AI features is not just a legal imperative — it’s a commitment to building trustworthy, fair, and human‑centric AI that respects individual rights.