EU AI Act Article 50 Transparency Assessor — Deployment Guide

See CHANGELOG.md for version history.

Overview

EU AI Act Article 50 Transparency Assessor — a standalone-but-suite-aware skill that identifies which
of the Art. 50(1)–(5) transparency duties apply to a system and guides what must be implemented and by
when. It produces two deliverables: a formal mini-report and a per-obligation compliance checklist
with gap flags.

Five duties, two roles — 50(1) interaction disclosure and 50(2) synthetic-content marking (provider);
50(3) emotion/biometric notice and 50(4) deepfake/public-interest-text labelling (deployer); 50(5)
delivery quality (cross-cutting)
Trigger + exemption logic — the average-consumer obviousness test (50(1)), the assistive-function
exemption (50(2)), the Art. 5 gate (50(3)), and the narrow 50(4) exceptions
Implementation depth — the final Code of Practice's layered marking architecture, the official EU
labelling icon set, and per-modality placement
Dated, Omnibus-aware roadmap — 2 Aug 2026, the 2 Dec 2026 legacy grace (adopted by Council 29 Jun 2026,
awaiting OJ), the 22 Jul 2026 signatory deadline, and the 2 Feb 2027 Code interoperability date
Standalone but chainable — ingests the classifier's ASSESSMENT CONTEXT block and emits its own
portable Art. 50 compliance block

File Structure

ai-act-transparency/
├── SKILL.md                              # Main skill instructions (deploy this)
├── CHANGELOG.md                          # Version history
├── evals/
│   └── evals.json                        # Test cases
└── references/
    ├── art50-duties.md                   # The five duties + 50(6) governance
    ├── obviousness-and-exceptions.md     # Obviousness test, exemptions, boundaries, cross-provision interactions
    ├── code-of-practice-final.md         # Final Code of Practice (10 Jun 2026) — provider marking + deployer labelling
    ├── commission-guidelines-art50.md    # Draft Commission Guidelines (8 May 2026)
    ├── eu-labelling-icons.md             # Official EU icon set + design/placement requirements
    ├── timeline-and-grace.md             # Dated roadmap + Digital Omnibus grace (adopted, awaiting OJ)
    ├── implementation-checklists.md      # Provider / deployer / SME action checklists
    ├── report-template-art50.md          # Mini-report, checklist, and portable compliance block templates
    └── sources.md                        # Audit-grade source manifest (URLs, status, last-checked, uncertainty tiers)

Deployment

Claude.ai (User Skills)

Go to Settings → Profile → Custom Skills (or equivalent)
Upload the entire ai-act-transparency/ folder structure
The skill auto-triggers on "Art. 50 transparency obligations", "do we need to label AI content /
deepfakes", "AI chatbot disclosure", "synthetic content marking", "Kennzeichnungspflicht", or
"Transparenzpflichten"

Claude Code / Custom MCP Setup

Copy the ai-act-transparency/ folder to your skills directory:

cp -r ai-act-transparency/ /path/to/your/skills/user/ai-act-transparency/

Ensure the skill is registered in your configuration

Usage

Quick Start

Either start fresh or hand over context from a prior skill:

"We're launching an AI support chatbot and an image generator under our own brand. What Article 50
transparency duties apply, what do we implement, and by when?"

Or chain from the classifier:

"Here's the ASSESSMENT CONTEXT block from the classifier — assess our Art. 50 transparency obligations
and produce the report and checklist."

Trigger Phrases

"Check Art. 50 transparency obligations" / "Transparenzpflichten"
"Do we need to label AI content / deepfakes" / "Kennzeichnungspflicht"
"AI chatbot disclosure" / "synthetic content marking" / "watermarking"
"What must we implement under Art. 50 and by when"

Workflow

Phase	Description
Phase 1: Intake	System description + optional `ASSESSMENT CONTEXT` ingestion
Phase 2: Role Determination	Provider / deployer / both
Phase 3: Trigger Determination	Per-duty trigger + obviousness/exception test
Phase 4: Implementation Deep-Dive	What to build per triggered duty
Phase 5: Dated Roadmap	Omnibus-aware deadlines
Phase 6: Output	Mini-report + checklist + portable compliance block

Regulatory Basis

Document	Reference
EU AI Act	Regulation (EU) 2024/1689, Article 50 + recitals 132–137
Deepfake definition	Art. 3(60)
Penalty band	Art. 99(4) — Tier 2 (EUR 15M / 3%)
Code of Practice on Transparency of AI-Generated Content	Final, 10 June 2026 (Art. 50(7))
Commission Guidelines on Art. 50	Draft, 8 May 2026 (Art. 96(1)(d))
Digital Omnibus	50(2) legacy-marking grace to 2 Dec 2026 — adopted (Council final green light 29 Jun 2026), awaiting OJ publication

License & Disclaimer

This skill produces structured Art. 50 transparency guidance based on Regulation (EU) 2024/1689, the final
Code of Practice on Transparency of AI-Generated Content, and the Commission's draft Art. 50 Guidelines. It
is not legal advice. The Code is voluntary and adherence is not conclusive evidence of compliance; only the
CJEU can authoritatively interpret Art. 50. Outputs should be reviewed by qualified legal counsel before
regulatory use.

Licensed under AGPL-3.0 — see LICENSE at the repo root.

Created by Oliver Schmidt-Prietz — OneZero Legal

eu-ai-act-transparency-assessor