Control · C-060 · Regulatory & Legal Compliance
Prohibited-practice screening
Every proposed AI use is checked against the practices the law prohibits before it is approved, the check is recorded, and it is repeated when the use changes.
- Control type
- Preventive
- Risks addressed
- 9
- Audience
- Heads-of · Practitioner
- Frameworks
- Article 5 · Article 5(ba) · Article 5(bb)
- Responsible AI domain
- Lifecycle stage
- Reviewed
- 2026-11-10
What it does
It stops a prohibited use before any money is spent on it. Each proposal is compared with a maintained list of banned practices, chiefly those in Article 5 of the EU AI Act, and with the organisation's own prohibited uses. A match ends the proposal; a near match goes to legal review. Compliance owns the criteria and the governance function applies them at intake and whenever a system's purpose or function changes.
How to implement it
- Maintain the list of prohibited practices from the current legal text, with plain-language questions for each.
- Add the organisation's own red lines from its risk appetite.
- Apply the questions at intake, to bought and built systems alike.
- Refer anything uncertain to legal, and record the reasoning.
- Screen again when a system gains a function, such as emotion inference or biometric analysis.
- Apply the same check to suppliers' products and features.
Evidence it operates
Design: screening questions tied to the legal provisions, with a review date. Operation: a screening record for each proposal; examples referred to legal; a proposal stopped or redesigned as a result; evidence that the list was updated when the law changed.
Where it fits
The first test at intake, before tiering. It is the primary control for the prohibited practices in this library, including AI-driven social scoring, workplace and education emotion recognition and subliminal or manipulative AI techniques. It is triggered by AI use-case intake and approval.
Limits
The prohibitions are drawn narrowly, with exceptions and conditions, so screening needs legal judgement and not only a checklist. A capability can be added after approval by a supplier's update. Screening catches what is proposed openly; it does not find a prohibited use that was never submitted. The response to a prohibition is not to build the system, so no later control compensates for a miss here.
Risks this mitigates
Primary mappings
- AI-driven social scoring Critical
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. An AI system evaluates or classifies people over time on their social behaviour or personal characteristics, and the score leads to detrimental treatment that is out of context or disproportionate.
- Biometric categorisation of sensitive attributes Critical
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. A biometric categorisation system categorises individuals from their biometric data to deduce or infer race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation.
- Exploitation of vulnerable groups by AI Critical
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. An AI system exploits vulnerabilities arising from a person's age, disability or social or economic situation to distort their behaviour in a way that causes, or is likely to cause, significant harm.
- Individual criminal risk profiling Critical
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. An AI system assesses or predicts the risk that an individual will commit a criminal offence, based solely on profiling them or on their personality traits and characteristics.
- Non-consensual intimate imagery and child sexual abuse material generation Critical
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. A system is offered or used to generate intimate depictions of identifiable real people without their consent, or child sexual abuse material.
- Real-time public biometric identification Critical
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. Real-time remote biometric identification is used in publicly accessible spaces for law enforcement, outside the three narrow objectives and the authorisation the law requires.
- Subliminal or manipulative AI techniques Critical
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. An AI system uses subliminal, purposefully manipulative or deceptive techniques that materially distort a person's behaviour and cause, or are likely to cause, significant harm.
- Untargeted facial recognition scraping Critical
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. An AI system creates or expands a facial recognition database through the untargeted scraping of facial images from the internet or from CCTV footage.
- Workplace and education emotion recognition High
- Screens the proposed use against the prohibition before approval. The response to a prohibition is not to build the system. An AI system infers the emotions of people at work or in education, for a purpose other than medical or safety reasons.
Framework and clause references
| Framework | Clause | Title |
|---|---|---|
| EU AI Act | Article 5 | Prohibited AI practices The prohibitions screened for. |
| EU AI Act | Article 5(ba) | Non-consensual intimate imagery The intimate imagery prohibition. |
| EU AI Act | Article 5(bb) | Child sexual abuse material The child sexual abuse material prohibition. |
Related terms
How to cite
Axiom Verity (2026). Prohibited-practice screening. Reference Library, C-060. https://axiom-verity.com/controls/prohibited-practice-screening. Reviewed 2026-11-10.
Found an error or a better source? Report a correction.