FIAT/IFTA World Conference 2026

The Images Sport AI Project is the second major phase of the IOC’s long‑term digital transformation of its archival ecosystem, following the successful migration from bespoke on‑premises systems to a cloud‑based, integrated architecture between 2021 and 2024. This foundation introduced seven interoperable platforms for preservation, access and analytics, and enabled the structured migration of more than 85,000 audiovisual assets and 928,000 photographs with full metadata. 
Launched in 2024 in alignment with the IOC AI Agenda, the Images Sport AI Project introduces state‑of‑the‑art artificial intelligence into both front‑end discovery and back‑end documentation workflows. Its dual objective is to enhance search capabilities in The Olympic Media Library (TOML) and accelerate metadata creation within the Media Asset Management system through automated tagging. 
Front‑end innovations include AI‑driven image and video similarity search, enabling users to retrieve visually related content that traditional text‑based descriptions cannot capture. A conversational search prototype extends this further by combining TOML metadata, multimodal search and IOC data, allowing natural‑language queries and richer discovery pathways. 
On the back end, automated tagging uses deep‑learning models are developed to generate sport‑specific vocabularies (actions, equipment, venues), general actions, logos, symbols and framing descriptors, while deliberately excluding athletes’ identities or emotions to ensure compliance with legal and ethical frameworks. Winter sports vocabularies form the basis for pilot deployments ahead of Milano Cortina 2026, with Summer sports following  for Dakar 2026 Youth Olympic Games and Los Angeles 2028 Olympic Games. 
The project directly addresses key questions about large‑scale infrastructure, automation, authority control and ethical metadata governance. It positions the IOC Archives as a global laboratory for innovation within Sport Archives, demonstrating how AI can strengthen discoverability, operational efficiency and long‑term stewardship across extensive heritage collections.

This paper presents the development of a high-speed video search system based on image features for a large-scale video archive, NHK Archives, without relying on manually created metadata. Conventional video retrieval systems typically require extensive manual metadata annotation. In contrast, the proposed system extracts still images from video clips and computes their features using the CLIP model, enabling multimodal search with text or images to retrieve semantically related scenes.
To achieve practical response times over a massive dataset, the system employs approximate nearest neighbor (ANN) search. In addition, a prototype function was developed to automatically generate program structures by combining speech recognition and large language models (LLMs), improving the interpretability of search results by clearly indicating the corresponding programs and scenes.
The system adopts a hybrid architecture in which computationally intensive processes—such as clip extraction, CLIP feature computation, and speech recognition—are executed on-premise, while structure generation and search operations are performed in the cloud. This approach significantly enhances the accessibility of video content while reducing the cost of metadata creation for large-scale video archives.

We present an initial methodological framework for representing events in the automated cataloguing of journalistic audiovisual content. Integrated into a broader research project on the automation of document analysis in newsrooms, this study examines how eventbased metadata can be effectively modelled using controlled vocabularies, without requiring structural modifications to existing MAM indexing
systems. The approach operationalizes the 5W model—adapted here as 4W + 1H—through lightweight, nonontological solutions suitable for legacy documentation environments.


The research draws on a corpus of more than 5,000 news items related to the 2024 DANA weather event in the Valencian Community. Using a representative sample, the study analyses three types of documentary elements—documents, segments, and stories—to identify common metadata patterns, assess their potential for reuse in cascading automated workflows, and determine the structural limits of storylevel aggregation in news production contexts.


The results reveal persistent issues such as overindexing, multisemantic fields, weak traceability, and inconsistent responses to the 5W framework. The study examines whether advanced semanticweb structures (ontologies, knowledge graphs) are necessary, or whether enhancing existing controlled vocabularies and refining metadata field architectures can provide sufficient robustness and operational value.


The findings underscore the need to improve metadata structures whose semantics depend both on their formal design and on their functional placement within the document record, as key factors for enhancing retrieval, consistency, and automation in audiovisual news cataloguing.

Artificial intelligence is rapidly becoming a primary interface to knowledge and culture. Audiovisual archives therefore cannot remain passive observers. They steward some of the richest records of public life: broadcast history, oral testimony, performance, news, memory and cultural expression. As AI systems increasingly mediate discovery, interpretation and reuse, archives face shared challenges around extraction, invisibility, misrepresentation, governance and reciprocity.
This panel frames these challenges through the concept of Public AI: an approach to AI that serves the public interest rather than purely commercial logics. For audiovisual heritage, this means three concrete things. First, public assets: archival data, metadata, tools and standards should remain as open, interoperable and reusable as possible, reducing dependence on closed platforms. Second, public purposes: AI should support access, interpretation, education, participation and creative reuse of audiovisual heritage, rather than merely mining archives as raw material. Third, public oversight: decisions about how archival content is used in AI, under what conditions, and with what safeguards, must remain subject to meaningful public accountability.
Together, these principles offer a practical alternative to extractive AI models. They position audiovisual archives not just as data providers, but as institutions that can shape more democratic and trustworthy AI futures.
In this panel, three speakers will explore how audiovisual archives can contribute to Public AI through data stewardship, contextualisation, standards, rights-aware access and collaborative governance. The discussion builds on the Alignment Assembly “Culture for AI” [1] and follow-up work linked to the Data Space for Cultural Heritage and the European Collaborative Cloud for Cultural Heritage.
Participants will gain concrete insights into how audiovisual archives can respond strategically to AI, including practical perspectives on metadata, rights, access, standards and collaboration that are directly relevant to the FIAT/IFTA community.
[1] https://www.dataspace-culturalheritage.eu/en/news/discover-our-new-publication-on-artificial-intelligence-and-the-data-space

Audiovisual archives are sovereign infrastructures of cultural memory. They are not merely collections of content: they are the evidentiary foundation of public discourse, the raw material of historical knowledge, and a shared resource whose integrity is a condition of trust in media institutions. Generative AI and deepfakes can pose a structural threat to this integrity — not only by enabling the production of false content, but by corroding the epistemic status of archival material itself. A manipulated archive is no longer a reliable source; it is a liability.
This paper argues that broadcasters occupy a distinctive position in the governance of AI-generated deepfakes: neither as technology providers nor as distribution platforms, but as custodians of audiovisual heritage with editorial responsibility, reputational capital, and existing technical capacity. From this position, broadcasters can act as upstream governance actors — intervening at the point of content production rather than relying exclusively on downstream detection, fact-checking, or user-labelling. The central claim is that voluntary adoption of content-marking standards (watermarking, C2PA/IPTC metadata, provenance certification) by broadcasters constitutes a form of shared technological capacity that strengthens the infrastructure sovereignty of audiovisual archives.
The paper develops its argument through three moves. First, it situates broadcasters within a polycentric governance framework (drawing on Ansell et al., 2023 and Floridi et al., 2024, 2026), distinguishing this approach from the normative inflation critique: the effectiveness of broadcaster-led governance derives not from adding further voluntary principles to an already saturated landscape, but from the specificity of institutional incentives and the concreteness of available technical tools. Second, it analyses Article 50 of the EU AI Act and the draft Code of Practice on AI-generated content transparency (currently under finalisation) as a case study of how regulatory frameworks interact with — and depend upon — horizontal governance by media institutions. Third, it addresses the question of transferability: drawing on critical perspectives, the paper asks what conditions must be met for broadcaster-led marking standards to function as genuine shared capacity across diverse institutional and infrastructural contexts.The argument is grounded in a body of emerging literature on AI governance, media self-regulation, and archival provenance standards, and engages directly with initiatives such as the EBU's sovereign AI framework, and the guidelines developed by the Archival Producers Alliance. 

When Reuters News Agency acquired the commercial archive licensing operation Screenocean, it inherited two distinct media asset management (MAM) systems. Maintaining these parallel environments — each shaped by its own metadata schemes, cataloguing practices, and operational workflows — introduced growing risks to preservation, discoverability, rights management, and long‑term sustainability. 
The challenge was twofold: migrating a large, active audiovisual archive while preserving the integrity and context of its assets. This paper outlines the methodologies used to address this challenge, including metadata mapping, risk assessment, and workflow redesign. 
This presentation examines the unification of these MAM environments as a case study in preservation through migration. Although driven by technical necessity, the consolidation was equally an exercise in ensuring continuity of archival stewardship amid organisational and technological change. 
Beyond large-scale asset migration, the paper explores effective approaches to MAM consolidation, including metadata harmonisation, duplication management, and the preservation of contextual and rights information that give archival materials their meaning and value. 
The project offers a practical case study for archives facing merger, restructuring, or infrastructure renewal, highlighting key lessons related to preservation, access, and sustainability. 

The rapid obsolescence of analog media and playback equipment represents a critical risk to the preservation of audiovisual archives. This paper presents a large-scale digitization project of Betacam tapes, starting with an initial scope of 50,000 media items, developed as a strategy to mitigate this risk and enhance the value of the collection.
To meet these demands, the FlyPack solution (by company Meta Martis) was adopted, enabling greater efficiency in the digitization workflow, optimized resource usage, and quality control within a high-volume environment. The proposal addresses the technical and operational challenges involved in the mass conversion of analog content to digital formats.
As a result, the project is expected to enrich the archive and expand opportunities for access, use, and reuse of the content. Digitization also paves the way for the application of artificial intelligence, enabling new forms of analysis, organization, and value creation, while fostering future opportunities within the audiovisual ecosystem.

The objective of this presentation is to reflect upon the state of digital preservation in Brazil within both the public and private sectors. This analysis is conducted through two case studies: Empresa Brasil de Comunicação (EBC) and Rede Globo. This work summarizes research compiled for an academic article currently in its final stages of preparation.
Established in 2007, EBC represents the unification of what is estimated to be nine public communication outlets, including TV Brasil. In addition to the television archives produced by the station itself, EBC inherited collections from defunct educational broadcasters: TVE Maranhão, founded in 1969, TVE Rio de Janeiro, created in 1975, and TV Nacional de Brasília, which was inaugurated in 1960. Over its nearly twenty-year history, the company has sought to adapt to technological innovations in the preservation of and access to television archives. Since 2015, EBC has utilized a Media Asset Management (MAM) platform, which enables the digital archiving of raw footage and television programs, as well as the viewing, downloading, and keyword-based searching of assets. This system is gradually replacing the legacy IAcervo platform, which primarily recorded media formats and metadata such as cataloging details, synopses, and production dates. In line with this modernization, the Archives and Research Department implemented a digitization project in 2019 for U-matic (KA) media. These tapes contain some of TVE Rio’s most significant programming, such as Sem Censura, No Mundo da Bola, and Edição Nacional.
Rede Globo, Brazil’s largest broadcaster, was founded in 1965. Its documentation efforts began in 1974 in Rio de Janeiro, though the CEDOC (Documentation Center) department was not officially inaugurated until 1976. The collection is stored on a wide variety of media formats developed over the decades for audio and video storage, ranging from Quadruplex and U-matic tapes to contemporary LTO (Linear Tape-Open) technology. Content that was originally captured on magnetic tape is now captured digitally via binary code; however, this still necessitates storage solutions, whether physical media or virtual environments such as the cloud. Rede Globo has also undergone a significant evolution in its processes, progressing from paper-based cataloging stored in folders to sophisticated computerized systems. Initially, these systems were developed by in-house engineers; subsequently, the company entered into agreements with Dalet, and later Vizrt and DIVA, to outsource these services. Concurrent with this process is the digitization of analog media, generating digital assets that allow for a fully digital production workflow. This transition further facilitates the long-term conservation of original physical documents by reducing the need for repeated handling, as digital versions become the primary point of access. The purpose of this thirty-minute presentation is, therefore, to reflect on the Brazilian preservation landscape through a dialogue between these two distinct institutional experiences.