NVIDIA and the EBU widen their sovereign AI framework to cover agentic security in media

NVIDIA used a creative-industry event in Berlin on October 19, hosted with Adobe, Bertelsmann and UFA, to lay out two media-focused announcements. The first is that the European Broadcasting Union is expanding the sovereign standards framework it built with NVIDIA last year for European public service media. The second is that Adobe's Frame.io is speeding up video agent search with NVIDIA GPUs, a capability already in use at German film and TV producer UFA.

According to NVIDIA's account, the expanded initiative now brings in sovereign cloud providers, open model developers and systems integrators. The goal is to define, in concrete terms, where broadcasters' agent workloads run, how their data is governed, which models power services, and how those systems stay secure through their lifecycle. That moves the effort from a statement of principles toward requirements and reference architectures.

The scale of the customer base explains the attention. The EBU describes itself as the world's leading alliance of public service media, with more than 115 member organizations in over 55 countries reaching more than 1 billion people. As agents move into newsrooms, production and audience services, those broadcasters want a foundation that protects editorial independence, data sovereignty and national resilience.

The division of labor is specific. Microsoft contributes expertise from Microsoft Sovereign Cloud, including Azure Local and Foundry Local, so broadcasters can process sensitive editorial material locally and keep critical services running when external connectivity is down. Nebius offers a platform for broadcasters to train, fine-tune and run their own models on European infrastructure. NVIDIA's Nemotron open models provide a base that broadcasters can customize for local languages, archives and editorial policies, while Qvest and other integrators supply media-specific consulting to make the reference architectures deployable in real broadcast environments.

"These standards give EBU members a shared foundation to strengthen resilience, retain control and deliver accountable advanced services that sustain audience trust," said Antonio Arcidiacono, the EBU's chief technology and innovation officer. Delphine Ernotte, CEO of France Televisions and president of the EBU, framed the same point around sovereignty: sovereign cloud infrastructure and open models can help Europe's public service media innovate while protecting data, editorial independence and cultural diversity.

The second announcement was more tangible. Adobe's GPU-accelerated search pipeline uses NVIDIA NVDEC for video decoding, TensorRT for optimized inference and the CUDA platform for preprocessing and embedding generation, with NVENC and NVDEC also accelerating media ingest. NVIDIA reports about 5x faster embedding generation, up to 4x higher throughput and up to 5x lower latency for search processing. A new two-frame-per-second proxy, roughly 80% smaller than the 360p version it replaces, cuts storage and transfer while keeping the visual detail search needs.

UFA shows why that matters in practice. The producer delivers more than 3,500 hours of programming a year, including the daily drama "Gute Zeiten, schlechte Zeiten." When editors assemble a recap or flashback, they may need a specific reaction or unresolved exchange from as far back as 15 years, something file names and folders cannot surface. Frame.io semantic search finds moments by concept, action or setting, and hybrid search adds notes, transcripts and approval status, so an editor can look for "a character looking concerned," narrow by character name and filter to approved footage.

The framework is less about keeping agents out of the newsroom than about deciding, in advance, whose cloud they run on, whose models they use and who can audit them. That makes public broadcasters an unusually demanding test case for sovereign agentic AI, and it puts pressure on vendors to prove that open models can pass scrutiny on local languages, archive compliance and auditability. NVIDIA CEO Jensen Huang's GTC Berlin keynote on October 21 is the next checkpoint for how far the reference architectures have actually progressed.