EU AI Act Prohibited Practices: The 8 Banned AI Uses Businesses Must Avoid in 2026
Most coverage of the EU AI Act focuses on the high-risk paperwork that phases in through 2026 and 2027. But the part that's already in force — and carries the steepest fines in the law — is the outright ban on certain AI practices. If your product does any of these eight things, no compliance documentation makes it legal.
Why the Prohibited Tier Matters More Than the High-Risk Tier
The EU AI Act sorts AI into four risk tiers: minimal, limited (transparency obligations), high (the bulk of the compliance burden), and unacceptable. The unacceptable tier is small but absolute — these systems are simply banned. There is no conformity assessment, no CE marking, no risk-management file that makes them legal. You either stop the practice or you stay out of the EU market.
Two facts make this the tier businesses should check first. It was the first part of the Act to apply (2 February 2025), and it carries the highest penalties in the regulation: up to €35 million or 7% of total worldwide annual turnover — higher than GDPR's ceiling. A feature you shipped without legal review could already be in breach.
The 8 Prohibited AI Practices (Article 5)
1. Subliminal, manipulative, or deceptive techniques
AI that deploys subliminal or purposefully manipulative/deceptive techniques to materially distort behavior in a way that causes (or is likely to cause) significant harm. This is the catch-all most consumer products need to read carefully — dark-pattern AI that nudges users into decisions against their interest can fall here.
2. Exploiting vulnerabilities
AI that exploits vulnerabilities of a person or group due to age, disability, or a specific social or economic situation, to materially distort behavior in a harmful way. Targeting financially distressed users or children with manipulative AI is the textbook example.
3. Social scoring
AI used by public or private actors to evaluate or classify people based on social behavior or personal characteristics, where the resulting score leads to detrimental treatment in unrelated contexts or that is unjustified/disproportionate. This is not limited to governments — private social-scoring systems are covered.
4. Predictive policing of individuals
AI that assesses or predicts the risk of a person committing a criminal offense based solely on profiling or personality traits. Risk-scoring individuals for future crime is banned (location-based crime forecasting and clearly evidence-based investigative tools are treated differently).
5. Untargeted facial-recognition scraping
AI that creates or expands facial-recognition databases through untargeted scraping of facial images from the internet or CCTV footage. This provision directly targets Clearview-style data collection.
6. Emotion recognition at work and in education
AI that infers emotions of people in the workplace or in education institutions — except for medical or safety reasons. Productivity/sentiment tools that read employee emotion are squarely in scope.
7. Biometric categorisation of sensitive traits
AI that categorises people based on biometric data to infer race, political opinions, trade-union membership, religious or philosophical beliefs, sex life, or sexual orientation. Narrow law-enforcement exceptions apply, but commercial use is banned.
8. Real-time remote biometric identification in public (law enforcement)
Real-time remote biometric identification in publicly accessible spaces for law-enforcement purposes is prohibited, subject to narrow, authorised exceptions (e.g. specific serious-crime scenarios). Most businesses won't touch this, but vendors selling to public-sector clients must.
The Practices Businesses Actually Trip Over
Most SaaS and consumer products will never build predictive-policing or remote-biometric systems. The realistic exposure for ordinary businesses comes from three of the eight:
Manipulative or deceptive AI (dark patterns)
HighIf your AI-driven UI, chatbot, or recommendation engine is designed to push users toward choices that harm them — hidden subscriptions, exploitative upsells, emotionally manipulative retention flows — you risk falling under the manipulation ban. Review any 'persuasion optimization' or AI-driven dark-pattern features.
Emotion recognition in HR / workforce tools
CriticalSentiment analysis of employees, AI that scores interview candidates on emotional cues, or 'engagement' tools that infer worker mood are banned in the workplace. This is the single most overlooked prohibition for B2B software vendors. Strip emotion inference from any product used by EU employers.
Private social scoring
HighIf your platform aggregates unrelated behavioral signals into a 'trust score' or 'reputation score' that then affects how people are treated in unrelated contexts, you may be running a prohibited social-scoring system — even as a private company. Tenant, gig-worker, and marketplace reputation systems should be reviewed.
A 5-Minute Self-Check
Before worrying about high-risk documentation, confirm you're not in the banned tier. Answer these honestly for every AI feature that touches EU users:
- Does it infer emotions of employees or students? If yes, and it's not for medical/safety reasons, stop it.
- Does it manipulate users into harmful decisions? Audit persuasion/retention AI for dark patterns.
- Does it build a cross-context "social" or "trust" score? Scope it to its original, justified purpose.
- Does it categorise people by biometric data to infer sensitive traits? Remove sensitive-trait inference.
- Does it scrape facial images to build a recognition database? This is banned — do not.
Find your compliance gaps before regulators do
RatedWithAI helps SaaS teams surface compliance and trust issues across their web properties. Start with a free scan to see how your product presents to users — and where your highest-risk surfaces are.
Scan Your Product for Free →Frequently Asked Questions
Is the prohibited-practices ban already enforceable?
Yes. The Article 5 prohibitions applied from 2 February 2025. National market-surveillance authorities and the EU AI Office can act on them now. Unlike the high-risk obligations that phase in later, there is no grace period for banned practices.
We're a US company — do we really need to care?
If your AI is used in the EU — sold to EU customers, deployed for EU users, or its outputs are used in the EU — you are in scope. The Act is explicitly extraterritorial, like GDPR. A banned feature reaching even a small number of EU users creates exposure.
What's the difference between 'prohibited' and 'high-risk'?
High-risk systems (e.g. AI in hiring, credit, critical infrastructure) are allowed but heavily regulated — risk management, data governance, human oversight, conformity assessment. Prohibited systems are not allowed at all. The compliance question for the banned tier is binary: stop the practice or exit the market.
Could a normal recommendation engine be 'manipulative'?
A standard recommendation engine is generally fine. The ban targets systems designed to materially distort behavior in a way likely to cause significant harm — exploitative, deceptive, or subliminal techniques. Ordinary personalization isn't prohibited, but persuasion features engineered to override user judgment to their detriment should be reviewed with counsel.