EU AI Act & AI Governance Toolkit — curated by Devseis
Collection
AI Act, GDPR/PII and LLM-security resources curated by Devseis, a Luxembourg AI governance & cybersecurity consultancy. Not legal advice. • 13 items • Updated
Datasets:
category stringclasses 6
values | domain stringlengths 9 64 | use_case stringlengths 22 175 | legal_basis stringlengths 7 30 | rationale stringlengths 20 92 | key_obligations stringlengths 16 218 |
|---|---|---|---|---|---|
prohibited | Manipulation | AI using subliminal or purposefully manipulative/deceptive techniques to materially distort a person's behaviour causing significant harm | Art. 5(1)(a) | Circumvents informed decision-making via covert techniques with material harm. | Banned outright — no compliance path, must not be deployed in the EU. |
prohibited | Exploitation of vulnerabilities | AI exploiting vulnerabilities due to age, disability, or a specific social/economic situation to distort behaviour causing significant harm | Art. 5(1)(b) | Targets known vulnerable groups to their detriment. | Banned outright. |
prohibited | Social scoring | General-purpose social scoring of individuals by public or private actors leading to detrimental or unfavourable treatment unrelated to the context in which data was generated | Art. 5(1)(c) | Cross-context behavioural scoring with punitive real-world consequences. | Banned outright. |
prohibited | Predictive policing | AI assessing the risk of an individual committing a criminal offence based solely on profiling or personality traits | Art. 5(1)(d) | Pure profiling-based individual crime prediction without objective, verifiable facts. | Banned outright (narrow law-enforcement support tools that add to a human assessment based on objective facts may fall outside this ban — case-by-case legal analysis required). |
prohibited | Biometric database scraping | Untargeted scraping of facial images from the internet or CCTV to build/expand a facial recognition database | Art. 5(1)(e) | Mass, non-consensual biometric data collection with no targeting. | Banned outright. |
prohibited | Emotion recognition at work/school | Inferring emotions in the workplace or education institutions | Art. 5(1)(f) | High risk of misuse in power-imbalanced settings. | Banned, except narrow medical or safety-reason exceptions. |
prohibited | Biometric categorisation (sensitive traits) | Categorising individuals by biometric data to infer race, political opinions, trade union membership, religious/philosophical beliefs, sex life or sexual orientation | Art. 5(1)(g) | Infers special-category data from biometrics. | Banned, narrow exception for lawfully labelling/filtering lawfully acquired biometric datasets or law-enforcement content moderation. |
prohibited | Real-time remote biometric ID in public spaces (law enforcement) | Real-time remote biometric identification in publicly accessible spaces for law enforcement purposes | Art. 5(1)(h) | Mass surveillance risk in public spaces. | Banned by default; narrow, judicially-authorised exceptions (e.g. victim search, specific terrorist threat, serious crime suspect search). |
high_risk | Biometrics | Remote biometric identification systems (non-real-time / post-event) | AI 2024/1689 Annex III(1) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Biometrics | Biometric categorisation systems inferring sensitive or protected attributes (where not prohibited) | Annex III(1) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Biometrics | Emotion recognition systems outside the prohibited work/education contexts | Annex III(1) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Critical infrastructure | Safety component managing/operating critical digital infrastructure, road traffic, or water/gas/heating/electricity supply | Annex III(2) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Education | AI determining access/admission to educational institutions | Annex III(3)(a) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Education | AI evaluating learning outcomes, including those used to steer a student's learning process | Annex III(3)(b) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Education | AI assessing the appropriate level of education for an individual | Annex III(3)(c) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Education | AI monitoring/detecting prohibited behaviour of students during tests | Annex III(3)(d) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Employment | AI used for recruitment or selection of natural persons, including targeted job ads and screening applications | Annex III(4)(a) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Employment | AI making decisions on promotion/termination, allocating tasks based on personality/characteristics, or monitoring/evaluating performance and behaviour | Annex III(4)(b) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Essential services | AI evaluating eligibility for essential public assistance benefits and services | Annex III(5)(a) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Essential services | AI evaluating creditworthiness or establishing credit scores (except fraud-detection use) | Annex III(5)(b) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Essential services | AI for risk assessment and pricing of life and health insurance | Annex III(5)(c) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Essential services | AI evaluating/classifying emergency calls or dispatching emergency first-response services | Annex III(5)(d) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Law enforcement | AI assessing an individual's risk of offending or reoffending, or risk for potential victims | Annex III(6)(a) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Law enforcement | Polygraphs and similar tools used by law enforcement | Annex III(6)(b) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Law enforcement | AI evaluating evidence reliability in criminal investigations/prosecutions | Annex III(6)(c) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Migration/border | AI used as polygraphs or to assess risk posed by a natural person entering/having entered a Member State | Annex III(7)(a) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Migration/border | AI examining asylum, visa or residence permit applications and associated complaints | Annex III(7)(b) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Justice/democracy | AI assisting judicial authorities in researching/interpreting facts and applying the law | Annex III(8)(a) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
high_risk | Justice/democracy | AI intended to influence the outcome of an election/referendum or voting behaviour | Annex III(8)(b) | Falls within an Annex III use-case category referenced by Article 6(2). | Risk management system, data governance, technical documentation, logging, transparency to deployers, human oversight, accuracy/robustness/cybersecurity, conformity assessment, EU database registration (Art. 8-15, 49). |
limited_risk | Conversational AI | Chatbot or voice assistant interacting directly with natural persons | Art. 50(1) | Interacts with or affects natural persons in a way that materially benefits from disclosure. | Must disclose to the person that they are interacting with an AI system, unless obvious from context. |
limited_risk | Synthetic content | AI generating or manipulating image, audio or video content (deepfakes) | Art. 50(4) | Interacts with or affects natural persons in a way that materially benefits from disclosure. | Content must be marked as AI-generated/manipulated in a machine-readable, detectable way. |
limited_risk | Synthetic text | AI generating or manipulating text published to inform the public on matters of public interest | Art. 50(4) | Interacts with or affects natural persons in a way that materially benefits from disclosure. | Must disclose the text was AI-generated, unless human-reviewed/edited with editorial responsibility. |
limited_risk | Emotion/biometric (non-prohibited use) | Emotion recognition or biometric categorisation system deployed outside prohibited contexts | Art. 50(3) | Interacts with or affects natural persons in a way that materially benefits from disclosure. | Must inform the natural persons exposed to the system of its operation; process personal data per GDPR. |
minimal_risk | Productivity | AI-powered spam filter | No Annex III/Art. 5/50 trigger | Not high-risk, prohibited, or subject to Art. 50 transparency duties. | No AI Act-specific obligations; voluntary codes of conduct (Art. 95) encouraged. Other law (GDPR, sector rules) may still apply. |
minimal_risk | Entertainment | AI-enabled non-player-character behaviour in video games | No Annex III/Art. 5/50 trigger | Same — minimal risk. | No AI Act-specific obligations; voluntary codes of conduct (Art. 95) encouraged. Other law (GDPR, sector rules) may still apply. |
minimal_risk | Operations | AI-based inventory or demand-forecasting system with no Annex III use-case | No Annex III/Art. 5/50 trigger | Same — minimal risk. | No AI Act-specific obligations; voluntary codes of conduct (Art. 95) encouraged. Other law (GDPR, sector rules) may still apply. |
minimal_risk | Customer support | Internal AI ticket-routing/classification tool not interacting directly with the public | No Annex III/Art. 5/50 trigger | Same — minimal risk, though GDPR still applies if personal data is processed. | No AI Act-specific obligations; voluntary codes of conduct (Art. 95) encouraged. Other law (GDPR, sector rules) may still apply. |
gpai | Foundation models | Provider placing a general-purpose AI model on the EU market | Art. 53 | GPAI providers have obligations regardless of downstream use-case risk tier. | Technical documentation, training-data summary, copyright policy, information to downstream providers. |
gpai_systemic_risk | Foundation models | GPAI model classified as having systemic risk (e.g. by training compute >10^25 FLOPs, or Commission designation) | Art. 51, 55 | Additional risk from scale/capability of the model itself. | Model evaluation & adversarial testing, systemic risk assessment/mitigation, incident reporting, cybersecurity protections. |
EU AI Act Risk Classification Reference
See also: the training and test data for the Devseis AI Act Classifier v5 is in
Devseis/devseis-ai-act-classifier-v5-data. This repo is the reference table the tier definitions came from.
A structured reference table mapping AI system use-cases to their EU AI Act (Regulation (EU) 2024/1689) risk tier — prohibited, high-risk, limited-risk (transparency), minimal-risk, or general-purpose AI (GPAI) — with the relevant Article/Annex citation and a summary of key obligations.
Compiled by Devseis, a Luxembourg cybersecurity, data protection and AI governance consultancy, from the Regulation's own text (Article 5 prohibited practices, Annex III high-risk categories, Article 50 transparency obligations, Articles 51/53/55 GPAI obligations).
⚠️ Not legal advice
This is a reference starting point, not a compliance determination. The AI Act's Annex III scope has been, and continues to be, refined by European Commission guidance and delegated/implementing acts. Always verify against the current consolidated text and, for an actual system, get a proper legal/compliance assessment — this is exactly the kind of judgment call that depends on your specific system and deployment context.
Columns
| Column | Description |
|---|---|
category |
prohibited / high_risk / limited_risk / minimal_risk / gpai / gpai_systemic_risk |
domain |
Sector or use-case family |
use_case |
Description of the AI system/use-case |
legal_basis |
Article/Annex citation |
rationale |
Why it falls in that tier |
key_obligations |
What the Act requires for that tier |
Used in
This reference table was the ground truth behind a follow-on experiment:
Devseis/eu-ai-act-classifier-experiment
tested whether a small classifier could learn these categories from
examples grounded in this table (and later the Regulation's official
text and Commission guidance directly). It didn't work reliably — see
that repo for the full, honest results. A live, two-model comparison
built on that experiment is at
Devseis/Caveat.
License
CC-BY-4.0 — compiled from public regulatory text.
- Downloads last month
- 57