Avoiding Prohibited AI Practices (Article 5 AI Act)

As of February 2, 2025, the EU has enforced a strict ban on certain AI applications: anyone who uses them regardless risks the highest fine in the entire AI Act. This affects not only tech corporations, but every company that uses or procures AI systems.

In brief: Article 5 AI Act bans certain AI practices EU-wide, such as social scoring, manipulative systems, systems that exploit people's vulnerabilities, and unlawful biometric evaluations. This affects every company that deploys or procures AI systems, regardless of industry or size. Violations of this ban carry fines of up to €35 million or 7% of worldwide annual turnover, the highest fine tier in the entire AI Act.

Who is affected by the ban?

The ban under Article 5 AI Act applies to every company that develops, procures, or deploys AI systems in its business operations, regardless of whether it is a large corporation or a small business. This covers not only in-house software, but also purchased AI tools, for example for recruiting, customer scoring, marketing personalization, or internal monitoring. Using ready-made AI tools from third-party providers does not exempt a company from this duty to check either, because ultimate responsibility lies with whoever actually deploys the system in their own operations, not only with whoever developed it.

Which AI practices are specifically prohibited?

Article 5 AI Act names several categories of prohibited practices. These include social scoring systems that evaluate people based on their behavior or characteristics and then disadvantage them in contexts unrelated to the original purpose. Also banned are AI systems that deliberately manipulate, deceive, or subconsciously influence people, as well as applications that knowingly exploit the vulnerabilities of specific groups, for example due to age, disability, or economic situation, in order to steer their behavior. This also covers unlawful biometric evaluations, for instance emotion recognition in the workplace or automated categorization of people based on sensitive characteristics.

What exactly do you need to do?

The legislator does not require a new documentation duty in the strict sense, but an active and recurring review of the AI systems you use and plan to use. In concrete terms, that means:

1. Systematically check AI systems in use and planned against the catalog of prohibited practices in Article 5 AI Act.
2. Assess particularly sensitive applications, such as those used to evaluate people or for emotion and biometric recognition, critically and, when in doubt, with expert support.
3. Consistently disable prohibited functions, or refrain from procuring such systems in the first place.
4. Bind the procurement and use of new AI tools to a mandatory approval check beforehand, instead of letting departments introduce tools unchecked.
5. Document the outcome of every review so that, if in doubt, you can provide evidence to supervisory authorities.

A tool like Company Audit shows which of these obligations specifically apply to your company.

By when do you need to act?

The ban on prohibited AI practices has already applied since February 2, 2025, making it the longest-standing obligation in the entire AI Act. Unlike many other AI obligations, this is not a one-time implementation deadline but an ongoing, permanent duty: every newly introduced AI system, or one whose function has changed, must be checked again against the list of prohibited practices, even if it was previously classified as uncritical.

What happens if you violate this ban?

Violations of Article 5 AI Act carry the highest fine tier in the entire AI Act: up to €35 million or 7% of worldwide annual turnover, whichever amount is higher. This puts the range well above the fines for other breaches of duty under the AI Act, such as violations of transparency or labeling obligations. This high threshold shows how seriously the legislator takes protecting people from manipulative and discriminatory AI practices.

Frequently asked questions

Are ordinary business AI tools such as chatbots or text generators affected?

Most standard tools for text creation, translation, or internal processes are uncritical. The main risk lies with systems that evaluate or monitor people, or that are designed to deliberately influence their behavior.

Is it enough to ask the AI provider whether their product is compliant?

No. The duty to check lies with the deploying company itself; a statement from the provider does not replace your own assessment against the list of prohibited practices, because the specific use within your own operations can be assessed differently than the general product description.

What counts as social scoring under Article 5 AI Act?

This refers to systems that evaluate people based on their behavior or characteristics and then disadvantage them in a context unrelated to the original purpose for which the data was collected.

Do I have to have every new AI tool checked in advance?

Yes. An approval check before procuring or using a new AI tool prevents prohibited functions from entering your operations unnoticed and unchecked.

What if I am unsure whether a system is affected?

Since the highest fine tier in the entire AI Act applies here, you should seek legal advice before using a questionable system if in doubt, rather than relying on a later clarification.

Source: Article 5 AI Act. This article provides general information and does not replace legal advice for individual cases. As of: July 2026.