/18AI Transparency & Content Marking
A plain-English guide to Article 50 transparency duties, machine-readable marking, disclosure, and why a label is not the same thing as provenance evidence.
Beginner
9 min
Open +
AI Transparency & Content Marking
A plain-English guide to Article 50 transparency duties, machine-readable marking, disclosure, and why a label is not the same thing as provenance evidence.
Beginner
9 min
Transparency rules are about helping people know when AI is involved. Some providers must technically mark synthetic outputs so they are machine-readable and detectable. Some deployers must visibly disclose certain deepfakes or public-interest text. A marker helps communicate origin. It is not a universal proof of everything that happened before or after publication.
Leave knowing / Separate disclosure, marking and evidence
Article 50
There are two different ideas hiding inside “AI transparency.”
For providers of AI systems that generate or manipulate synthetic audio, image, video or text, Article 50 includes machine-readable marking and detectability duties in the cases covered by the law, subject to its scope and exceptions.
For deployers, Article 50 also contains disclosure duties for certain uses, including deepfakes and certain AI-generated or manipulated text published to inform the public on matters of public interest. These transparency obligations apply from 2 August 2026, with a limited transition for some pre-existing systems under the updated rules.
Those duties are context-specific. A watermark, metadata field, label or machine-readable signal does not automatically prove who generated the content, which workflow produced it, whether it was later edited, or whether every legal duty was satisfied.
A disclosure tells people something. Evidence helps you show what happened.
Three layers
Do not collapse these into one feature.
Disclosure
A person is told that AI was involved or content was artificially generated or manipulated.
Marking
A technical signal is attached or embedded so synthetic content can be detected in a machine-readable way where required and feasible.
Evidence
A separate record helps establish what system activity happened at runtime and gives a later reviewer something to verify.
Before publishing
Ask the boring questions.
Is this output synthetic audio, image, video or text?
Are we the provider of the generating system or the deployer using it?
Could the content be mistaken for authentic real-world media?
Is it a deepfake within the meaning of the Act?
Is the text being published to inform the public on a matter of public interest?
Does an exception or human editorial responsibility change the disclosure analysis?
If a marker disappears later, do we still have a separate runtime record of what happened?
What each layer can prove
Be precise about the claim.
| Signal | Useful for | Do not assume it proves |
|---|---|---|
| Visible AI label | Human disclosure | Who created it, exact generation event, legal compliance |
| Machine-readable marker | Automated detection of synthetic origin where supported | That the marker survived every transform or platform |
| Runtime evidence record | Showing a recorded system event and later matching evidence | A legal conclusion, human authorship, or compliance certificate |
Official sources
We simplify the map here. For a real legal decision, use the law, current guidance and qualified counsel.
Where Paper Trail fits
Available now / Not a compliance certificate
If the question is “what happened?”, keep the receipt.
Paper Trail is being built to keep bounded runtime evidence around supported AI-system activity so a team can inspect what happened later. It does not decide your AI Act role, classify your system, or certify compliance.
Paper Trail is still a private preview. Only capabilities that are actually available are presented as available.
Notes / Subscribe
Get the next note.
No schedule. No content calendar. A note goes out only when there is something worth understanding.