article-51 July 2026DILAIG

Our Free Tool to Detect Prohibited AI Practices (Article 5)

Article 5 of the EU AI Act has banned 8 AI practices since 2 February 2025. Our free tool helps you identify in minutes whether your system is affected — and what to do if it is.

Last updated: July 2026 · Reading time: 6 minutes

Article 5 of the EU AI Act came into force on 2 February 2025 — well ahead of the high-risk AI obligations (August 2026). It outright bans eight categories of AI practices. No compliance pathway, no technical documentation that makes them lawful, no CE marking process. These systems are simply illegal in the EU.

The problem: many organisations don't know whether they're affected. The prohibited practices are defined in technical legal language, and the boundary between what is banned and what isn't is often unclear in practice.

That's why we built the DILAIG Article 5 Detection Tool.

What the tool does

The tool is a deterministic decision tree covering all 9 prohibitions under Article 5. It asks you targeted questions about your system — in 5 to 10 minutes — and tells you whether one or more prohibited practices are potentially at issue.

It covers all 9 prohibitions:

#	Prohibited Practice
1	Subliminal manipulation (Art. 5(1)(a))
2	Exploitation of vulnerabilities (Art. 5(1)(b))
3	Social scoring by public authorities (Art. 5(1)(c))
4	Criminal risk assessment based on profiling (Art. 5(1)(d))
5	Untargeted scraping for biometric databases (Art. 5(1)(e))
6	Emotion recognition in workplaces and schools (Art. 5(1)(f))
7	Biometric categorisation by sensitive characteristics (Art. 5(1)(g))
8	Real-time remote biometric identification in public spaces (Art. 5(1)(h))
9	Non-consensual synthetic intimate content and CSAM — Art. 5(1)(i), Digital Omnibus 2026, applicable from 2 Dec. 2026

For each prohibition, the tree also maps the legal exceptions defined in the regulation — including the narrow medical and safety exceptions for emotion recognition, and the three specific cases where real-time RBI may be authorised by a Member State.

Why a deterministic tree, not an LLM

We deliberately chose not to use a language model for this tool. When the question is "is your system banned by law?", the answer cannot be probabilistic.

A deterministic decision tree guarantees:

The same questions in the same order for every user
No hallucinations about exceptions or legal thresholds
A traceable result your legal team can audit

LLMs remain useful for generating documentation, synthesising policies, or explaining regulatory concepts. Not for high-stakes regulatory screening.

How to use the tool

The tool is available for free, without creating an account, at Article 5 — Prohibited Practices.

What you need before you start:

A functional description of your AI system (what it does, not how it's built)
The deployment context: industry, end users, primary use case
If your system analyses natural persons: the nature of input data

The tool asks yes/no questions about your system. At each step, a "Why this question?" box explains the legal logic behind it — so you understand what you're assessing, not just what you're answering.

At the end, you receive:

A clear verdict for each prohibition: Affected / Not affected / Exception may apply
For each "Affected" result: the exact article text, a plain-language explanation, and recommended next steps
An exportable PDF summary

What the tool does not do

The tool does not replace legal advice. It is designed to help you quickly identify whether a situation warrants further scrutiny — not to provide a definitive ruling on the lawfulness of your system.

Specifically:

It does not cover high-risk AI obligations (Annex III)
It does not generate the required regulatory documentation (Annex IV, FRIA, EU Declaration of Conformity)
It does not substitute for legal analysis on borderline cases

For those steps, the full DILAIG audit takes over.

Typical use cases

Startups in product development Before investing in an AI feature involving user analysis — facial recognition, behavioural analysis, profiling — the tool lets you quickly check whether the concept itself is viable from a regulatory standpoint.

Legal and compliance teams When there is uncertainty about an existing system, the tool structures the analysis and identifies which prohibition is potentially at issue before engaging external counsel.

M&A due diligence When acquiring a company that develops or uses AI, the tool provides a documented Article 5 first-pass screening that can be integrated into the due diligence file.

DPOs and compliance officers Article 5 has been in force since February 2025. For organisations that have not yet formally reviewed their AI systems, the tool provides a structured starting point.

What to do if the tool detects a prohibited practice

If the result indicates your system matches a practice prohibited by Article 5, the immediate steps are:

Do not wait: the prohibition has been in force since 2 February 2025. Each day of continued use constitutes an ongoing violation.
Consult a specialist lawyer: penalties for Article 5 violations are the highest in the AI Act — up to €35 million or 7% of global annual turnover.
Assess whether a system modification is possible: in some cases, removing the prohibited component (for example, stripping the emotion inference layer while retaining other functions) can bring the system into compliance. This depends on the specific architecture and requires legal analysis.
Document the decisions taken: even if you immediately cease the prohibited practice, retain a record of when the decision was made and what measures were taken.

FAQ

Is the tool's output legally binding? No. It is a decision-support tool, not legal advice. The results are designed to structure your analysis and identify questions to raise with your counsel — not to replace that counsel.

Does the tool cover the Article 5(2) exceptions for real-time RBI? Yes. For Prohibition 8 (real-time remote biometric identification), the tool incorporates the three exceptions provided by Article 5(2) and the associated procedural conditions (judicial authorisation, spatial and temporal limitation, notification to the supervisory authority).

My system uses emotion recognition but only for safety purposes — is it prohibited? Article 5(1)(f) explicitly provides an exception for medical and safety uses. The tool will guide you in assessing whether your use case falls within this exception. The key boundary is intent: detecting operator fatigue to prevent an accident (safety) is different from analysing employees' emotions for HR performance management.

How long does the questionnaire take? Between 5 and 10 minutes, depending on the complexity of your system.

Get started

→ Access the Article 5 tool for free

To go further — full audit, regulatory documentation generation (FRIA, Annex IV, EU Declaration of Conformity):

See all DILAIG features → · Pricing →