EU AI Act Article 5: prohibited AI practices

Article 5 of the EU AI Act establishes a set of outright prohibitions on AI systems that pose unacceptable risks to people's safety, rights, or democratic values. Unlike the obligations in other parts of the Act, these bans are not subject to a transition period. They came into force on 2 February 2025 and apply immediately to any provider or deployer operating in the EU.

If your product or service uses any of the systems described below, you are already in scope. This is not a future obligation to prepare for — it is a current legal requirement.

What does Article 5 actually prohibit?

The EU AI Act Article 5 bans six categories of AI practice:

1. Subliminal and manipulative AI techniques

AI systems that deploy subliminal techniques beyond a person's consciousness, or that exploit psychological weaknesses or biases, to materially distort a person's behaviour in a way that causes or is likely to cause harm are prohibited. This targets systems designed to manipulate users without their awareness, nudging them into decisions they would not otherwise make.

2. Exploitation of vulnerable groups

AI systems that exploit specific vulnerabilities of persons due to their age, disability, or social or economic situation to distort their behaviour in a way that causes or is likely to cause harm are banned. This covers predatory targeting of elderly users, people with cognitive disabilities, or those in financial distress.

3. Social scoring by public authorities

Public authorities (and entities acting on their behalf) are prohibited from using AI to evaluate or classify individuals based on their social behaviour over time, or on their predicted personal characteristics, and then treating them in detrimental ways that are unrelated to the context in which the data was generated. This is the AI Act's explicit response to state-level social credit systems.

4. Real-time remote biometric identification in publicly accessible spaces

The use of real-time remote biometric identification systems in publicly accessible spaces for law enforcement purposes is banned, with narrow, strictly defined exceptions. The exceptions require prior judicial or independent administrative authorisation and apply only to specific serious crimes (such as terrorism, trafficking, or murder), targeted searches for missing persons, or preventing imminent threats to life. Retrospective use of such systems is permitted only for the prosecution of serious criminal offences and with appropriate authorisation.

5. Biometric categorisation by sensitive characteristics

AI systems that categorise individuals based on biometric data to infer their race, political opinions, trade union membership, religious or philosophical beliefs, sex life, or sexual orientation are prohibited. Note that biometric verification systems (which confirm a person's identity rather than classify them) are treated differently and are not covered by this ban.

6. Emotion recognition in workplaces and educational institutions

AI systems designed to infer the emotions of natural persons in the context of workplaces and educational institutions are banned. There are narrow exceptions for safety-relevant purposes, such as detecting driver fatigue. Outside those exceptions, employers and schools cannot lawfully use emotion recognition AI on employees or students.

Who does Article 5 apply to?

These prohibitions apply to both providers (companies that develop or place AI systems on the market) and deployers (companies or public bodies that use AI systems in their operations). Geographic scope follows the same logic as the rest of the Act: if the output of your AI system affects people in the EU, you are in scope regardless of where your company is based.

Private sector companies, public authorities, and law enforcement bodies are all covered, though law enforcement has specific conditional exceptions for biometric identification as noted above.

If you are a SaaS provider, a startup building an AI product, or a company integrating third-party AI tools, you need to understand whether any component of your stack falls within these banned categories. This includes AI features embedded in otherwise ordinary products, such as an HR platform that analyses employee facial expressions during video calls, or a consumer app that uses biometric data to infer religious affiliation.

For a broader view of who the Act applies to, see our guide: do I need to comply with the EU AI Act?

What must providers and deployers do?

The obligation is straightforward in principle: do not develop, place on the market, put into service, or use any AI system that falls within the prohibited categories.

In practice, this requires:

• Inventory your AI use cases. Identify every AI system your organisation develops or uses, including third-party tools and APIs. Document what each system does, what data it processes, and what decisions or outputs it produces.
• Screen against the Article 5 categories. For each system, assess whether it could reasonably be characterised as performing one of the banned functions. Pay particular attention to systems that process biometric data, infer personal characteristics, or are used in HR, security, or customer-facing contexts.
• Retire or redesign non-compliant systems. If a system falls within a prohibited category, it must be discontinued or fundamentally redesigned so that it no longer performs the banned function. There is no risk mitigation path for prohibited AI, unlike the requirements for high-risk AI systems, which can be compliant if managed correctly.
• Document your assessment. Even where you conclude a system is not prohibited, maintain a written record of that analysis. Regulators will expect to see evidence of due diligence.
• Review third-party contracts. If you are a deployer using a provider's AI system, review your contracts to understand how liability is allocated and whether your vendor has conducted their own Article 5 assessment.

Key deadlines

Article 5 prohibitions have been in force since 2 February 2025. There is no grace period.

Other EU AI Act obligations come into effect on different timelines:

• 2 August 2026: GPAI model obligations (fines of up to EUR 15 million or 3% of global turnover) and Article 50 transparency obligations (AI-generated content labelling) take effect.
• 2 December 2027: Annex III high-risk AI system obligations apply to existing systems placed on the market before the Act's entry into force.

For a full breakdown of penalties, see our guide on EU AI Act fines and penalties.

What to do now

If you have not yet assessed your AI systems against Article 5, start immediately. The prohibitions are already law. A regulator investigating a complaint, or auditing your organisation, will expect to find evidence that you have considered whether your systems are lawful.

Prioritise your review based on the types of data your systems process and the contexts in which they operate. Biometric data, workplace AI, and consumer-facing AI that could be characterised as manipulative are the highest-risk areas.

Where you identify systems that are potentially prohibited, take legal advice before continuing to operate them. The Article 5 bans carry the highest penalties in the Act: up to EUR 35 million or 7% of global annual turnover, whichever is higher.

Use the free ActComply risk screener to check your obligations in under five minutes: https://www.getactcomply.com/check