Kilde › Guides › AI Act Art 50 › Machine-readable marking under 50(2): …
Current to 26 August 2026 · updates land in the changelog.
The marking duty is an engineering obligation wearing legal clothes. Article 50(2) wants synthetic audio, image, video and text marked machine-readably and detectably — “effective, interoperable, robust and reliable as far as technically feasible”, a qualifier that both acknowledges reality and scales with provider size.
The Transparency Code of Practice (adequacy-endorsed July 2026 — signatories can rely on its measures to demonstrate compliance for 50(2), (4) and (5)) reflects the deployed state of the art as layers: primary — signed, tamper-evident metadata, where C2PA is effectively the only broadly deployed technology; secondary — imperceptible watermarking; optional — fingerprinting or registries. No single technology is mandated; the layering is the point, because each layer fails differently.
Images are mature: major generators ship C2PA credentials, verification tooling exists. Audio and video: workable via the same metadata layer plus watermarks. Text is the weak modality — no reliable, interoperable text-marking standard is deployed; vendor text-watermarks have been announced (one major provider's announcement explicitly concerned future models, with detection tooling “soon”) — and an announced capability is not a live one until verified against what your pipeline actually emits. For text, the technically-feasible qualifier plus documented best efforts is the defensible position, and first C2PA-conformant text-marking tools only began certifying in mid-2026.
Marks die in transit: re-encoding, thumbnailing, platform ingestion and screenshot-and-repost all strip metadata, and watermark robustness varies by transformation. Upstream marks from your model vendor do not clearly discharge the system provider's duty — after the final guidelines the duty is yours as the app provider — so the evidence that matters is a survival test matrix: each output modality × each real delivery path × what survives. That matrix, dated, is the artifact a regulator or enterprise buyer can actually accept.
The AI Act Article 50 Kit: the disclosure copy library (English + 中文), the marking implementation guide with survival-test protocol, deepfake and text-labeling walkthroughs, obviousness and scope memos, the evidence log, the Code of Practice accession path — built from the regulation and the final guidelines, with pinpoint citations.
Get the kit — US$190 Free 4-page sample (PDF)Instant download · 14-day unconditional refund · single-organisation licence · full product page
General information only — not legal advice, and no clearance opinions. Sources are Regulation (EU) 2024/1689 (Articles 50 and 99), Regulation (EU) 2026/1744, the Commission's final Article 50 guidelines of 20 July 2026 (non-binding) and our audited kit research. Marking-technology status changes fast: treat vendor announcements as live only once verified. © 2026 Kilde.
Built by Kilde's founder, a practising attorney admitted to a US state bar (not an EU or Hong Kong admission). About · Verification log · Refunds · Terms · Privacy · esau@trykilde.com