NVIDIA, EBU expand sovereign AI framework for broadcasters and media
NVIDIA said on Monday that it is widening its work on agentic AI security and media workflows, with the European Broadcasting Union expanding a sovereign framework for broadcasters and Adobe Frame.io using NVIDIA GPUs to speed video search for production teams. The announcements, made at a creative industry event in Berlin hosted with Adobe, Bertelsmann, NVIDIA and UFA, centered on data control, model choice and faster access to archived footage.
EBU broadens sovereign AI framework
NVIDIA said the EBU is extending the sovereign standards it introduced last year for European public service media. The expanded effort now includes sovereign cloud providers, open model developers and systems integrators to define how broadcasters can deploy agents while retaining control over data, infrastructure, models and security. The EBU said the framework is intended to support editorial independence, data sovereignty and national resilience as agents move into newsrooms, production and audience services.
Antonio Arcidiacono, chief technology and innovation officer at the EBU, said the standards give members "a shared foundation to strengthen resilience, retain control and deliver accountable advanced services that sustain audience trust." Delphine Ernotte, CEO of France Télévisions and president of the EBU, said sovereign cloud infrastructure and open models can help public service media innovate while protecting data, editorial independence and cultural diversity.
NVIDIA said Microsoft will map the standards to sovereign deployment options through Microsoft Sovereign Cloud, including Azure Local and Foundry Local, to support local processing of sensitive editorial material and continuity when external connectivity is unavailable. Nebius will provide a platform for broadcasters to train, fine-tune and run models on European infrastructure. NVIDIA also said its Nemotron open models are part of the framework, alongside consulting and systems integration from Qvest and other partners.
Adobe Frame.io cites faster search and lower storage needs
Adobe said it is using NVIDIA accelerated computing to improve video search in Frame.io. According to the company, the GPU-accelerated pipeline uses NVIDIA NVDEC for video decoding, TensorRT for AI inference and CUDA for preprocessing and embedding generation. Adobe also said GPU-accelerated ingest uses NVIDIA NVENC and NVDEC to speed encoding and decoding across larger media libraries.
Adobe said the system delivers about 5x faster embedding generation, up to 4x higher throughput and up to 5x lower latency for search processing. It also uses a new two-frame-per-second proxy that is about 80% smaller than the 360p version it replaces, reducing storage and transfer requirements while preserving visual detail needed for search.
The company said the workflow is relevant for UFA, one of Germany’s leading film and TV producers, which delivers more than 3,500 hours of programming a year, including the daily drama "Gute Zeiten, schlechte Zeiten." Adobe said Frame.io semantic search can find moments by concept, action or setting, while hybrid search can combine notes, transcripts and approval status. In the example given, an editor can search for "a character looking concerned," narrow by character name and filter to approved footage before moving into review.
Security context and limitations
The Berlin announcements focus on infrastructure and workflow controls for media organizations, not on a standalone product called "agentic security for media." NVIDIA’s broader technical materials describe agent security in terms of behavioral safeguards and infrastructure controls, including least-privilege access, sandboxing, logging and human approval for writes to systems of record. Those descriptions are NVIDIA’s own and do not constitute independent verification that the controls prevent all unauthorized activity.
Separately, NVIDIA’s recent security research posts on open-weight models, prompt injection and continuous agent monitoring provide context for the company’s broader agent-security push, but the source material supplied for this article does not establish a direct operational deployment of those tools in the media announcements made in Berlin.