eu-ai-act-compliance

🤖 EU AI Act Compliance Skill for Claude

An EU AI Act knowledge base as a Claude skill — so your AI assistant flags regulatory risks before you ship.

Regulation (EU) 2024/1689 · Updated March 2026

---

What is this?

An open source Claude skill that gives Claude deep knowledge of the EU AI Act — the world's first comprehensive AI regulation.

Once installed, Claude will proactively warn you about AI Act requirements when you're building, deploying or discussing AI systems in the EU. It knows what's already in effect, what's coming, and what's prohibited.

Examples of when the skill activates:

• You're building an AI-powered hiring tool (Annex III high-risk)
• You're deploying a chatbot without transparency disclosure (Art. 50)
• You're using emotion recognition in a workplace context (Art. 5 — prohibited)
• You're building on top of a GPAI model like an LLM (Ch. V obligations)
• You're discussing whether your AI system needs a conformity assessment

Why this matters right now

The AI Act isn't a future concern — parts are already in effect:

Date	Status	What applies
2025-02-02	IN EFFECT	Definitions, scope, AI literacy
2025-08-02	IN EFFECT	Prohibited practices, GPAI rules, fines up to 7% turnover
2026-08-02	UPCOMING	High-risk obligations, deployer obligations, transparency
2027-08-02	FUTURE	AI in regulated products

The 8 prohibited AI practices (Art. 5) and GPAI obligations are enforceable today.

Contents

eu-ai-act-compliance/
├── .claude-plugin/
│   └── plugin.json                    ← Plugin manifest
├── skills/eu-ai-act-compliance/
│   ├── SKILL.md                       ← Instructions for Claude
│   └── references/
│       └── eu_ai_act_en.json          ← Full structured knowledge base
├── README.md                          ← You are here
└── LICENSE                            ← MIT

What the knowledge base covers

Topic	Reference	Highlights
Phased timeline	Art. 113	What applies now vs. 2026 vs. 2027
8 prohibited practices	Art. 5	IN EFFECT — facial scraping, emotion recognition, social scoring etc.
Risk classification	Art. 6	4-tier system + decision tree
Annex III high-risk	Art. 6(2)	8 categories from biometrics to justice
High-risk requirements	Art. 8-15	Risk management, data governance, human oversight, documentation
Deployer obligations	Art. 26	Logging, informing users, incident reporting
FRIA	Art. 27	Fundamental Rights Impact Assessment
Transparency	Art. 50	Chatbots, deepfakes, generative AI, biometrics
GPAI models	Ch. V	Documentation, training data, copyright — IN EFFECT
Systemic risk GPAI	Art. 55	FLOPs > 10^25, adversarial testing — IN EFFECT
Fines	Art. 99	Up to EUR 35M / 7% for prohibitions — IN EFFECT
Danish supervision	—	Digitaliseringsstyrelsen + Datatilsynet split
GDPR interplay	—	10-step screening sequence + 7 key overlaps

Companion skill: GDPR Compliance

The AI Act and GDPR apply cumulatively. This skill cross-references the GDPR Compliance Skill for data protection matters. Install both for complete EU AI regulatory coverage.

Installation

Claude Desktop / Claude Code

1. Download or clone this repo
2. Place the eu-ai-act-compliance/ folder in your skills directory
3. Claude will automatically draw on AI Act knowledge when the conversation requires it

Manual use

Upload SKILL.md and references/eu_ai_act_en.json directly into a Claude conversation for ad hoc use.

Quick start

Once installed, type /euaiact in any conversation to see the skill status, enforcement timeline, last update date, and available commands. Claude will also activate automatically whenever it detects AI Act-relevant context.

What the skill does in practice

Art. 5 screening — Claude flags prohibited practices immediately:

• "This emotion recognition use in a workplace context is prohibited under Art. 5(f) — this applies now."

Risk classification — Claude walks you through the decision tree:

• "Your AI-based credit scoring system falls in Annex III category 5 — that's high-risk. Obligations apply from August 2026, but you should start compliance design now."

Timeline awareness — Claude distinguishes between what applies now and what's coming:

• "GPAI documentation requirements (Art. 53) are already in effect. The deployer obligations (Art. 26) kick in August 2026."

Disclaimer & Limitation of Liability