In a public broadcaster producing everything from breaking news and radio to streaming drama and multimedia-enriched articles, metadata is created everywhere — but accountability for it is rarely defined. Descriptions flow into the archive from many directions: through integrations with planning systems and publication platforms, through manual updates from spreadsheets, and through workflows the archive sometimes only discovers after the fact. Each route carries its own assumptions about quality, completeness, and meaning — assumptions made by teams describing content for their own immediate needs, not for its long-term life. As AI enters production and distribution workflows, the question of who governs media data is no longer academic: it determines whether content remains findable, reusable, rights-cleared, and trustworthy — and for a public broadcaster, whether it can honour its mandate as custodian of audiovisual heritage. 
This talk shares insights from an ongoing data governance exercise at VRT, approached deliberately from the archive’s vantage point. We examine three intertwined questions. First, where do the responsibilities of the archive end and those of production and data teams begin — and what happens in the gaps between them? Second, how do we establish traceability in an era of AI-generated and AI-modified content, where provenance is both harder to capture and more critical than ever? Third, can shared ontologies — specifically EBUCorePlus — serve as a backbone for company-wide governance across radically different media types, or do they promise more interoperability than they deliver in practice? 
Rather than presenting a finished framework, this talk offers an honest look at the tensions, trade-offs, and early lessons of governing data in an organisation where a news clip, a podcast episode, and a fiction series must somehow live in the same information ecosystem — and where the archive may be the only team with a view across all of it.