AI Transparency
Check first. Then label clearly.
SKILL.md in the open agentskills.io standard — works directly in Claude, ChatGPT/Codex, Cursor, Copilot & more.
AI Transparency is an AI skill for AI content labelling, disclosure and machine marking that, for a given AI content, system, chatbot, campaign or platform post, works out whether a labelling duty can apply, **who** is responsible, **which** current rule governs it, whether the platform already offers a label — and whether that is enough. It covers the EU AI Act (Art. 50), DACH, international modules, platform functions and C2PA. Built responsible: source-first & verify-live, never "legally safe/guaranteed", not legal advice.
What this skill does
- Determines the **role** (provider/downstream/deployer/publisher/platform/agency/private) *before* the duty, with a one-line reason
- Separates the **four Art. 50 tracks**: interaction, machine-readable provider marking, emotion/biometrics, deepfake + public-interest text
- Keeps **visible disclosure, technical provider duty, national misleading risk, platform rule and voluntary best practice** on separate lines
- Mandatory **platform-capability check**: exact control, menu path, format, first exposure, persistence on download/cross-posting — and whether an own notice is still needed
- **Source-first & verify-live**: never states rules/deadlines/toggles from memory; checks official primary sources; no current source → no green result (fail-closed)
- Covers **EU AI Act, EU Code of Practice & EU icons, DACH** plus **China, South Korea, US (federal + California), UK/CA/AU/JP (watchlist)** with an honest support status
- Explains **C2PA/Content Credentials** plainly — without presenting it as a blanket compliance seal
- **Label generator**: short label + full notice + accessible alt-text + position + technical & platform variant (chatbot/image/audio/video/text/real-estate/art)
- Outputs an **AI Transparency Check** (result class A–H, confidence, reasoning, implementation, source status) and an optional closing **mockup**
- Escalates high-risk cases (real people, deepfakes, elections, biometrics, large campaigns, your own AI product) to individual legal review
Description
AI Transparency (German: KI-Transparenz) is an AI skill for the AI-labelling duty, disclosure and machine-readable marking — it doesn't just say "yes" or "no", it works out, traceably: **who** is responsible, **which** current rule applies, does it fit **this exact** content and country, does the **platform** already offer a suitable label, is the **toggle enough**, what must be added, **how** to word and place the notice, and how to document the decision. Its guiding idea is **role-based**: it determines the role (provider, downstream provider, deployer, publisher, platform, agency, private person) *before* the duty. It keeps the **four Article-50 tracks** strictly separate — direct interaction (chatbot disclosure), machine-readable provider marking of generated content, emotion recognition / biometric categorisation, deepfakes and certain public-interest texts — and never conflates **visible disclosure** with **technical provider duty**, or **law** with **platform rule**. Central is the mandatory **platform-capability check**: instead of blindly "write an AI notice", the skill first checks what the platform itself labels (YouTube, TikTok organic & ads, Meta/Threads, Google Ads, LinkedIn, Pinterest, X, Snapchat) — the exact control, menu path, format, whether the notice is visible at first exposure and whether it survives download or cross-posting — and adds only what is still missing. A platform label is **never** an automatic compliance proof. The skill is deliberately **responsible**, because wrong labelling advice can cause real harm: it never states rules, deadlines, thresholds, toggles or UI paths from memory as fact, treats its local base only as a dated baseline (2026-07-14) and verifies every decision-relevant statement **live against the current, ideally official source** (EU AI Act, Commission guidelines, EU Code of Practice, national laws, official platform help, the C2PA specification). If something can't be verified, it produces **no green result** (fail-closed) and states an honest source status. It **never** says "legally safe", "guaranteed" or "replaces a lawyer", reports a confidence level and open points, escalates high-risk cases (real people, deepfakes, elections, biometrics, minors, large campaigns, your own AI system) to an individual legal review, and is **not legal advice**. Beyond the EU and DACH it covers international modules (China, South Korea, US federal + California, UK/CA/AU/JP as a watchlist) with an honest support status, explains C2PA/Content Credentials without presenting it as a blanket seal, delivers a **label generator** (short label, full notice, accessible alt-text, position, technical and platform variant) for chatbot, image, audio, video, text, real-estate/product visualisation and art — and, once a check is complete, optionally offers a fitting example mockup. Every larger output is an AI Transparency Check with a result class (A–H), reasoning, separated duty lines, implementation, wording, open points and a source status.
Examples
What it does not do
- Gives **no legal advice** and never says "legally safe", "guaranteed no duty" or "replaces a lawyer" — names possible duties with source, date and confidence
- Never states rules/deadlines/thresholds/toggles/UI paths from memory as fact and invents no sources — verify-live, else fail-closed
- Doesn't merge the four Art. 50 tracks, and separates visible disclosure from machine-readable marking, and law from platform rule
- Doesn't turn "no law found" into "no duty" and doesn't output a draft as if it were in force
- Doesn't treat a platform toggle / EU icon / auto label as an automatic compliance proof and invents no platform UIs
- Doesn't assume one country — establishes role, target market and the EU output-in-the-Union rule and flags unsupported jurisdictions
- Doesn't conclusively decide copyright, personality rights, data protection or contracts — only flags them
Compatibility & tech
- Tested (internal)
- 7 scenarios
- Recommended runtime
- Stärkstes verfügbares Modell; für die Live-Prüfung von Recht und Plattform-Funktionen eines mit Websuche/Deep Research; für Content-Audits eines mit Datei-/Bild-Analyse.
- Modes
- Quick check (default) · Deep check (full decision tree) · Provider check (role & machine-readable marking) · Content audit (actually analyse image/video/audio/text) · Platform check (capability resolver per format & surface) · Code-of-Practice gap analysis · Label generator (label package) · Source update / Legal Update Delta
- Inputs
- Description of the AI content / system / chatbot in plain language · Role, target markets, language, publication context · File or URL (image/video/audio/text) to analyse · Target platform & format (organic, ad, own website, download) · The AI tool used (for C2PA/watermarks)
- Output format
- AI Transparency Check (result class A–H + confidence, markets checked, check date, role, why, separated duty lines, recommended implementation, suitable wording, not-fully-checked areas, sources, source status); for platform cases an additional platform-check block with a post-publish checklist; optionally a closing mockup. Native buttons/tables/checklists where available, numbered fallback otherwise.
- Subcategory
- AI compliance & transparency navigation
- License
- Proprietär
Security profile
Runs entirely on your machine with your own AI — no external runtime, no running costs.
Only instructions, templates and references — no executable scripts.
Accesses external sources / the network in normal use (e.g. live pages, search/data APIs).
What you get
- ai-transparency-1.0.0/14 files
- .claude-plugin/marketplace.json
plugins/ai-transparency/skills/ai-transparency/14 files
- SKILL.md
- manifest.json
references/12 files
- ai-transparency-checklist.md
- ai-transparency-method.md
- eu-article-50.md
- examples-and-tests.md
- international-modules.md
- label-library.md
- mockup-generator.md
- national-dach.md
- platform-capability.md
- roles-and-jurisdiction.md
- sources-and-freshness.md
- technical-c2pa.md
- .agents/skills/ai-transparency/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
- LICENSE.txt
Installation
Also works as a chat prompt
No AI tool? Paste it into Claude, ChatGPT or Gemini and use the method right away.
This skill ships no scripts, so the prompt carries its full method.
Works best when your chat has web access.
Installing is the full version — it triggers automatically, runs its scripts and loads references as needed. As a chat prompt you drive the method by hand.
Unlock to copy the ready-to-paste prompt — then in “My Skills”.
Reviews
No reviews yet – be the first.
Note
Non-binding, source-based orientation — not legal advice and no substitute for an individual legal review. The skill never says "legally safe" or "guaranteed" and treats its local base only as a dated baseline (2026-07-14): rules, deadlines and platform functions change — verify every decision-relevant statement on the current official source, and without such a source it produces no final result. For real people, deepfakes, elections/political ads, biometrics/emotion recognition, large campaigns, government use or your own commercial AI system, involve an individual legal review and the competent authority. The final Art. 50 guidelines, still pending at the baseline, must be re-checked before productive use. Research baseline of the bundled examples: 2026-07-14.
Changelog
- v1.0.014.07.2026[object Object]
Frequently asked questions
What does AI Transparency do?
AI Transparency is an AI skill for AI content labelling, disclosure and machine marking that, for a given AI content, system, chatbot, campaign or platform post, works out whether a labelling duty can apply, **who** is responsible, **which** current rule governs it, whether the platform already offers a label — and whether that is enough. It covers the EU AI Act (Art. 50), DACH, international modules, platform functions and C2PA. Built responsible: source-first & verify-live, never "legally safe/guaranteed", not legal advice.
Do I have to label AI-generated content — and when?
It depends on your role, the content and the country — which is exactly what AI Transparency works out. The skill first determines your role (provider, deployer, publisher, platform, private …), then checks the relevant rule (EU AI Act Art. 50, unfair-competition law, platform rule) and tells you whether a visible label, a machine-readable provider marking or no duty applies. It deliberately avoids a blanket yes/no and gives a reasoned, source-based classification with a confidence level.
Is AI Transparency legal advice, or does it guarantee compliance?
No. AI Transparency is a non-binding, source-based orientation, not legal advice, and never says 'legally safe' or 'guaranteed no duty'. It names the checked rules with source, date and status, flags open points and a confidence level, and escalates high-risk cases (real people, deepfakes, elections, biometrics, large campaigns, your own AI product) to an individual legal review. The competent authority or a lawyer decides.
Does ticking the YouTube or TikTok AI box cover the legal duty, or do I need more?
AI Transparency checks exactly that with a platform-capability check: does the platform (YouTube, TikTok organic/ads, Meta/Threads, Google Ads, LinkedIn, Pinterest, X, Snapchat) offer an official labelling function, for which format, where does the notice appear, is it visible at first exposure and does it survive download/cross-posting? Often the toggle is the right UI overlay but not an automatic compliance proof — the skill says when you must additionally embed your own visible notice.
How does AI Transparency know the current legal status?
The skill's local knowledge is only a structure, a search map and a dated baseline (2026-07-14) — not durable truth. AI Transparency is source-first and verify-live: every decision-relevant statement is checked against the current, ideally official primary source (EUR-Lex, the EU Commission, national laws, official platform help, the C2PA spec). If something can't be verified, it produces no green result (fail-closed) and states the source status instead.
What counts as a deepfake under Article 50 — only celebrities?
No. AI Transparency applies the EU AI Act's deepfake definition broadly: AI-generated or manipulated image/audio/video that resembles existing persons, objects, places, institutions, organisations or events and could falsely appear authentic. That includes cloned voices, an AI-furnished real flat, a fabricated corporate event or a realistic fire in a real building — not only faces or celebrities. Art/satire is no blanket exemption; it only limits the form of disclosure.
Does AI Transparency cover countries beyond the EU?
Yes. AI Transparency treats the EU/EEA, Germany, Austria and Switzerland as core modules and has international modules for China, South Korea, the US (federal + California) and a UK/Canada/Australia/Japan watchlist. Each module shows its support status, and the skill applies the EU rule that the AI Act can reach non-EU actors when the output is used in the Union. For several target markets it offers a common minimum label plus country-specific add-ons.
Which AI tools does AI Transparency work with?
Claude (Projects & system prompt) · ChatGPT (Custom GPT & Custom Instructions) · Any LLM · Strongest with web search / deep research / file & image uploads (verify official sources live) · Native UI (buttons/tables/checklists) where available, Markdown fallback otherwise
How do I use AI Transparency?
AI Transparency is a SKILL.md in the open agentskills.io standard: install it with one command (npx) or download it and add it to your AI tool — Claude (Projects), ChatGPT (Custom GPT), Cursor, Copilot, Gemini CLI and more. No code needed.