EU AI Act labeling for generated media

In late July 2026, Meta introduced “Content Seal,” an invisible watermark embedded in images and videos produced by its Muse AI models. The company also released a web tool for users to verify the watermark, positioning the technology as a transparency measure for its platforms.

Independent testing quickly highlighted serious shortcomings: the watermark failed to be detected in over half of the images after modest cropping, and it only identifies Meta‑generated content, lacking compatibility with existing standards such as Google’s SynthID or the open C2PA framework.

Simultaneously, the European Union’s AI Act set to take effect on 2 August 2026 will require all AI‑generated media to carry a clear, machine‑readable label. The regulation gives major platforms—including Meta’s Instagram and Facebook, TikTok, Google, and OpenAI—until 2 December 2026 to implement detection and watermarking solutions that meet the broader labeling mandate, with penalties for non‑compliance.

The juxtaposition of Meta’s proprietary watermark and the EU’s comprehensive labeling rules underscores the growing pressure on tech firms to adopt interoperable, reliable provenance tools for synthetic content as the continent moves toward stricter AI governance.