The EU AI Omnibus provisional agreement, reached in May, grandfathers AI systems already on the market before 2 August 2026, giving them until 2 December 2026 to meet the machine-readable marking requirement under Article 50 of the EU AI Act1. Article 50 requires synthetic output, AI-generated audio, image, video or text, to carry a marking a machine can detect. This corrects what Lowdown told readers in updates #3 and #4, where we framed 2 August as a single hard cliff for EU broadcasters. For tools already in use, the operative deadline is now 2 December, so the broadcasters we described as scrambling have four more months than we reported.
The 2 August date still bites in one place. Any new generative-AI deployment launched after it gets no grandfathering and faces the standard immediately. A broadcaster running ChatGPT-driven content discovery today has until December; the same broadcaster shipping a synthetic-presenter product in September does not. The urgency has not vanished, it has moved from installed systems to new launches.
Spotify read this early and adopted DDEX, the recorded-music metadata standard, in May to carry AI-provenance flags through its existing licensing pipes. That is the compliant-by-design route the grandfathering now gives other firms more time to copy: thread provenance through metadata you already send, rather than retrofit a marking layer under deadline pressure. The Code of Practice that operationalises Article 50 was due to finalise in early June; as of 10 June no EU broadcaster has publicly signed it, and Meta has declined to sign the related general-purpose AI code.
