What Article 50 actually requires — and how synthetic content must be made detectable in practice.
The EU AI Act introduces enforceable requirements for how synthetic content must be disclosed and identified.
Under Article 50, providers of AI systems that generate or manipulate content must ensure that outputs are:
clearly identifiable as artificial
machine-readable
reliably detectable
This shifts transparency from a communication issue to a technical requirement
In practice
AI-generated content must not only be labelled — it must be detectable beyond the original platform.
This applies to:
Key principle
Transparency must persist across systems, formats, and transformations.
Many organisations assume that compliance can be achieved through:
visual disclaimers
text labels
standard metadata
This is incorrect.
The core problem
When content is:
— metadata is often removed or corrupted.
Result
The compliance signal disappears.
Robustness
The AI Act does not just require disclosure — it requires that solutions are:
This creates a critical constraint:
If the signal does not survive distribution, it does not meet the requirement.
Without persistent marking:
platforms cannot reliably detect synthetic content
compliance cannot be verified
disclosure cannot be trusted
This creates downstream risk:
inability to prove compliance
increased exposure under enforcement
reduced trust in digital content
What's required
To meet AI Act transparency requirements, organisations must implement systems that ensure:
01
Synthetic content must carry a signal that remains intact across transformations.
02
The signal must be detectable by systems — not just visible to users.
03
The marking must survive sharing across platforms and environments.
04
It must be possible to demonstrate that content has been correctly labelled.
This is not a labelling problem — it is an infrastructure problem.
Current approaches fail because:
signals are tied to platforms, not content
metadata is not persistent
detection is unreliable across systems
documentation is incomplete
Result
Compliance exists in theory — but not in practice.
Broader system
AI transparency does not operate in isolation.
It directly impacts:
platform enforcement obligations (DSA)
content traceability and auditability
user trust and authenticity
SASHA's role
SASHA enables organisations to embed machine-readable, robust identification directly into digital content.
This allows:
synthetic content to remain detectable across transformations
platforms to identify and act on synthetic media automatically
compliance signals to survive distribution
organisations to generate verifiable audit trails
By combining pixel-level watermarking with cryptographically bound metadata (C2PA), SASHA ensures that transparency is not lost when content moves across systems.
This enables organisations to meet the AI Act's requirement for effective, robust, and reliable technical implementation.
The shift
AI transparency is no longer about informing users.
It is about enabling systems to:
detect
verify
act
on synthetic content.
This represents a shift from:
Labels
Metadata
Best effort
Persistent signals
Embedded marking
Verifiable compliance
The AI Act requires more than disclosure policies.It requires systems that ensure persistent identification, reliable detection, cross-platform consistency, and verifiable compliance.
