What does EU AI Act Article 50(2) require?

Q: What does EU AI Act Article 50(2) require?

Article 50(2) requires providers of AI systems generating synthetic content to ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, across text, audio, video, and image, subject to technical-feasibility exceptions.

EU AI Act Article 50(2): Machine-Readable Marking of AI Output

Article 50(2) of the EU AI Act requires that providers of AI systems generating synthetic content ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. This is not a watermarking mandate — it is a provenance mandate. The provision applies to text, audio, video, and image outputs from general-purpose AI systems, with technical feasibility exceptions.

As of August 2026, the implementing guidance is still in draft, but the direction is clear: vendors entering the EU market after the February 2026 phased effective dates must demonstrate marking capability. The Council of AI provides a verified measurement credential (DEFONEOS-SEAL) that measures a system's marking pipeline — not that it marks perfectly (no system does at 0% false-positive), but that the marking is auditable, measurable, and consistently applied. C2PA Content Credentials, IETF SCITT transparency ledgers, and ISO/IEC 22128-1 content authentication are the reference frameworks Council measurement tests against.

For procurement: if a vendor claims Art 50(2) compliance, ask for the Council measurement card showing their actual detection rates — not their marketing copy.

Standards and sources referenced

EU AI Act 2024/1689, Article 50(2) on transparency obligations for providers and deployers
Council of AI Containment Incident Index
CSOAI Verified Measurement Credential (DEFONEOS-SEAL)
C2PA Content Credentials Specification v2.1
IETF SCITT Architecture draft-ietf-scitt-architecture