IBC RED LOGO HERE
IBC 2026 ACCELERATOR MEDIA INNOVATION PROGRAMME
FRAMES: Federated Retrieval, Agentic Media Environments and Software-Defined Workflows
GREEN PAPER:
An R&D Exploration of AI Ethics, Responsibility, Transparency & Risk for Creative Media Production
Edition One [September 2026]
R&D material for FRAMES Final POC Results
AUTHORS:
IBC Lead/Editor: muki Kulhan, IBC Accelerators
Project Co-Leads & First Authors: Roberto Iacoviello (RAI), Paola Sunna (EBU), Chris Vienneau, Raymond Drury (MovieLabs)
Project Team- First Authors: Tim Deussen (XRBB- Extended Reality Berlin-Brandenburg), Kathryn Webb and Ahsan Mallick (AIMICI), Ben Schofield and Aaron Bhugobaun (GC-SC - The Global Creative & Security Community)
Project Team- Contributing Authors:
John Canning (AMD), Erik Weaver (ETC/USC- Entertainment Technology Center at University of Southern California), Florian Reimann(MovieLabs), Clive Santamaria (ITV), John Maxwell Hobbs (Streamline Media), Nick Kennedy (BBC), Nicolas Frei (Enigmatic), Shamir Alibhai (Press Play Labs)
TABLE of CONTENTS
PART TWO: Project Learnings & Insights 6
How do we stay in Compliance with the EU AI Act of 2026? 7
PART THREE: Creating best practice for applied Responsible AI Governance & transparency 9
Responsible AI best practice 9
The framework for AI transparency on productions 10
Framework in practice on FRAMES 11
In development / coming soon: An industry-standardised globally-recognised Responsible AI stamp 11
PART ONE:
Executive Summary
AI is already an integral part of production workflows in the media world, spanning everything from script development and screenwriting to production and post-production. However, the rapid adoption of AI brings with it complex ethical, legal, and governance challenges. The question is not whether to use it, but whether one can demonstrate that it is being used responsibly.
This paper, stemming from the work of the IBC Accelerator FRAMES, offers an overview of AI-driven content and media experiences, with a focus on ethics and accountability. It identifies the most pressing challenges and issues, providing insights and practical guidelines to help stakeholders implement AI responsibly within the media sector. Neglecting these aspects could put an organisation on the wrong side of a very expensive conversation.
Introduction
Media industries are leveraging AI to transform key stages of the media value chain, ranging from production to distribution and monetisation.
AI is being integrated throughout the entire pipeline, from scriptwriting and screenwriting to music composition, editing, and animation creation. Much of the AI skills and experience is concentrated in a small group of freelancers working on the frontier, AI skills are in short supply
However, these advancements also raise critical questions regarding authenticity, attribution, copyright, transparency, accountability, and ethical use These all present a new set of commercial risks that need to be quantified to drive executive focus, the mitigation will depend on the potential costs especially when considering use of IP and archive materials. It is becoming clear that these models and platforms present a new set of security risks that require a different approach to current controls and governance.
The situation is further complicated by industry dynamics, organisational structures, and the skill sets of human professionals.
Global regulatory frameworks differ significantly, requiring companies to adapt responsibly to regional requirements.
To embed responsibility into AI systems within the media sector, monitoring and evaluation efforts are guided by several key concerns and priorities: copyright, transparency, privacy and data protection, and evolving policies and regulations. This necessitates the development of shared industry best practices and ongoing stakeholder engagement.
Context and Scope
FRAMES is one of the eight projects selected for the IBC Accelerator 2026 program.
The acronym stands for “Federated Retrieval, Agentic Media Environments, and Software Defined Workflows.”
The project involves a team comprising major media industry organisations and vendors, including RAI, EBU, and MovieLabs as co-leads, alongside members such as XRBB, ETC-USC, ITV, Enigmatic, Yamdu, Eddie AI, Streamline, GC-SC, and AIMICI, and is supported by AMD, Google Cloud, and Shure.
Its goal is to create an end-to-end framework for media production based on AI agents and a structure provided by the “MovieLabs Ontology for Media Creation.” The project’s scope encompasses AI-assisted media production, focusing on pre-production and production phases, and involves key activities such as scriptwriting, character generation, shot and image creation, and background music composition, all essential for producing a final video of professional quality.
While the market offers numerous AI tools and agents, AI cannot become a genuine competitive advantage for an organisation without a robust framework covering data, skills, governance, and accountability. The scenarios analysed within the project provided a meaningful context for examining the challenges and considerations involved in defining a framework for responsible AI in the media sector.
Definitions
Within this paper, we will refer to the following terms - this is what they mean in the context of media production when AI is involved
Copyright - The legal right to claim creative ownership over a particular asset or piece of media, whether produced with AI or not. Specifically for the purposes of control, reproduction, distribution and display.
Transparency - Typically, transparency is discussed in the context of a record of what kind of data AI tools have been trained on. However, in this paper we are primarily focused on the transparency around AI usage and decision-making on the project - production-level AI transparency, not tool-level. To be transparent about this, a record must be created containing accurate information.
Disclosure - This is the action of explaining to external parties how AI has been used on a project. This is usually best supported by accurate records of transparency, although it can also be a broader agreed statement (eg. AI has been used on this project, no AI has been used on this project etc.)
Responsible AI - The broad practice of putting effort towards ensuring that AI is used within a production in a responsible way. Common themes include Copyright, data security etc.
Soloprenuer definition
EU AI Act
Key Project Drivers
The adoption of AI in the media production sector is driven by the need for innovation, productivity improvements, and enhanced human capabilities, whilst keeping creative integrity intact. With this in mind, and within their short five-month sprint, this IBC Accelerator team prioritised routinely exploring these aspects, while also asking questions about AI Ethics, Responsibility, Risk, and Transparency that remain fundamentally important in media production. These questions were explored and left for further exploration through an extension of this research and paper. These aspects include, but are not limited to, questions such as:
- How can legitimate ownership and copyright regarding AI-generated content be determined?
- If content is generated exclusively by artificial intelligence without sufficient human creative input, can it be legally protected under copyright law, or are competitors free to copy it directly?
Who is liable for due diligence and risk assessment, especially if something goes wrong? The prompt engineer? The production company?The technical team? The AI model provider or the creator platform? More importantly, at what stage of the production should this be considered?
What exactly are the guarantees of the legal indemnities? These are contractual safeguards designed to protect users against copyright infringement lawsuits arising from generative AI output or training data. However, the scope and applicability of such protections vary significantly depending on the training architecture and the provider’s terms of service. Under what conditions do such indemnities become completely void?
What rights do original authors retain when their works are included in training sets?
How can artists and publishers protect their works when billions of copyrighted assets are extracted without explicit consent, attribution, or financial compensation?
PART TWO: Project Learnings & Insights
Learnings
Copyright
- The biggest overall tension when designing the best human-AI hybrid workflow on the project was between achievable quality and ensuring copyright claimability.
- The only way to truly know the copyright-ability of a final piece of media and its content (eg. characters)
Transparency
- Substance
- Having a simple human-readable record explaining how AI has been used on the project is really helpful as an internal project reference, helping to build awareness and understanding of how and why tools have been considered within the process.
- It’s difficult to make judgements on how “important” or “significant” AI usage is for any particular use case without digging further into specific usage actions (inputs, outputs etc.).
- Process
- Although it is best to establish a process for transparency documentation from the start of a project, it is possible to achieve it mid-way through or even retroactively.
- On projects where communication happens mostly online,some decision-making and tool usage details can be captured from meeting transcripts and chat conversations, rather than relying solely on proactive self-reporting or deep technical documentation. Using AI tools responsibly to support this effort is also possible. However ultimately these approaches are still error-prone and information should always be reviewed and confirmed by the team before finalisation.
- Even if information can’t be documented in real time from start to end of a project, it is important to at least have a responsible person or people who know the project and process well enough to help capture the info retroactively with confidence
Other
- Tradeoffs were necessary between different responsible AI aspects of the project in order to achieve acceptable quality of the final film, in line with professional-grade animation standards. For example:
- Some AI tools have safeguards in place for flagging the use of identifiable copyrighted material as inputs, however, when you already own the rights to the content this kind of functionality is obstructive rather than useful. This can lead to choosing tools that are less protective to achieve (still legitimate) outputs
How do we stay in Compliance with the EU AI Act of 2026? Industry insights:
While larger studios continue to integrate AI technologies into their operations, a growing number of independent, small-to-midsize, and AI-native studios are emerging. These entities frequently operate independently of broader industry initiatives, executing productions with minimal centralized studio oversight while rapidly navigating evolving AI governance standards. Consequently, any ethical best practices and framework developments must account for sole proprietors and AI-native studios amid ongoing industry disruption. Furthermore, there is a clear need for a global industry initiative to establish an approved directory of AI tools, evaluating vendors on operational practices, commercial viability, ethical track record, and transparency.
PART THREE: Creating best practice for applied Responsible AI Governance & transparency
Responsible AI best practice
[Summary including reference to IBC last years paper, other responsible ai approaches]
AIMICI has been a partner to the project and also has been working on standards around responsible AI best practice in industry over the last two years. In their approach, they have identified two key levers to help the industry navigate AI concerns at scale:
Training
Not to be confused with AI training, this educational effort is about ensuring everyone working on productions, especially key decision-makers, have an understanding of what AI can do, it’s common risks, a baseline understanding of how AI governance and escalation processes work, and the role each of us play within it.
This is applied training that includes,
By ensuring this kind of training is included, not just skills or tools training, it builds a safer foundation for innovation as tools continue to progress and improve.
This has been established in industry so far via standardised responsible AI training courses that have been accredited by UK skills training body ScreenSkills, released in May 2026 and rolled out through organisations such as Pact.
Transparency
With the foundation of training in place, the issue of lack of transparency of AI usage across productions can now also be tackled fully. Productions must be able to accurately document and appropriately disclose the use of AI on their productions, internally and externally.
The framework for AI transparency on productions
A proposed new framework has been designed by AIMICI and tested on the project, focused on how to bring transparency and understanding of copyright implications to projects that have hybrid AI and human workflows such as this.
The transparency framework, in summary, looks to capture the following scope of attributes
- WHAT - Details of the project and it’s key phases
- WHO - Who is responsible for overseeing and reporting on AI usage across those stages
- HOW - What are the AI use cases and tools used
- WHEN - When were tools used and what decisions were made on which use cases or tools can proceed or be stopped
- WHERE - Where did AI actions take place across a workflow, and what assets were affected
The key to applied transparency is ensuring appropriate documentation and records are in place, so that appropriate disclosure is possible at different levels. The AIMICI framework proposes the following three levels of records:
| Level 1: Credits Statement - For audiences and press A short, plain-language statement in the credits (one to three individual sentences or statements) that the production creates according to the uses’ importance to the production and/or its compliance needs. |
|---|
| Level 2: Production AI card - For industry & stakeholders One-to-two-page reference detailing AI use cases and tools, and relevant attributes around risk, visibility and significance of the AI use. This is organised into production stages, and includes a high-level record of human oversight over the governance process, and the businesses involved in the record-keeping across the production. Also includes a more detailed set of disclosure statements that form the “long list” from which the Level 1 individual statements are chosen. This could be made publicly available as a website or hosted document. |
| Level 3: Auditable AI record - For commissioner-level governance & legal review Full record including signed declarations, decisions and high level actions log and further details about tools and licensing at a minimum. Can also include records of consent, and granular actions & asset-level data such as the machine-readable AI marks, which are required by the EU AI act. |
Framework in practice on FRAMES
Within the FRAMES project, there was an opportunity to create a production AI card / level 2 record for an AI-centric production process for the first time. We also explored how the team’s proposed workflow could feed into a version of Level 3 records. We explored these two approaches in tandem.
Outcomes:
- We were able to produce a reviewed production AI card
- Credit statement is still in progress (this is finalised at the end of the project based on finalise records)
- The auditable record has evolved throughout the project. With the EU AI act requirements coming into force during the project, we began to see some machine-readable AI marks emerge on some of the AI tools that were used. However ultimately we had to rely mostly on manual tagging of human vs AI-driven assets and workflows within the Streamline platform
In development / coming soon: An industry-standardised globally-recognised Responsible AI stamp
Agreeing a standardised format for a credit statement / on-screen disclosure is a must-do for the industry. However, a written statement alone still requires both legal sign off and intepretability by audiences who may still be at a lower level of AI literacy. For example, judging whether it is “good” or “bad” that a production has used AI to support VFX activities is not clean cut - the use cases and tools are the deciding factor on how responsibly this has been done.
Therefore, on top of a baseline statement, a further evolution has also been proposed by AIMICI - an easily understandable Responsible AI stamp that can be added to productions.
There have been attempts at this in a number of ways already across industry, (REFERENCES) however there has not yet been anything that considers responsible AI working practice at large on a production.
By adhering to the combined requirements of appropriately trained staff and transparency documentation, AIMICI’s new Responsible AI mark is intended to signify best practices across a range of productions including film & TV, scripted & unscripted, and AI-first, hybrid and traditional approaches.
Further details and consultation opportunities will open in Q4 2026. Learn more here:
PART FOUR:
Findings from the project in the context of the AI EU act
The “soloprenuer” in content production, actual best practice on the frontier for major studios, advertisers and content creators
There are a set of emerging standards that could help drive best practice in the use of AI in production. The rate of innovation in models, commercial pressures on traditional media distribution and advertising and the rapid inflows of investment capital have made this technology extremely fast moving. The baseline set of basic metadata including the identity of content, people, devices and software together with time and location that need to be captured to support proper attribution and audit of new workflows. Where possible this should rely on existing initiatives such as OpenTimeline, SMPTE RIS-OSVP, C2PA/CAWG. The experience gained through this project suggests some summary best practice guides that could help build new AI skills for current creative talent.
A closing word from Tim Deussen, XRBB
Five months of building FRAMES taught us that the hard problem in AI-assisted production is not generation. It is proof.
Almost every difficulty we met had the same shape. We could push for the quality a professional animation demands, or we could keep a clean claim to copyright, and holding both at once meant slowing down and writing things down. Machine-readable AI marks began appearing on our tools while the project was still running, but never on all of them, and never in a form our workflow could gather up, so in the end we tagged human and AI-driven work by hand. Safeguards designed to protect an anonymous user got in the way of a production that already owned the rights to its own material. And none of the record could be handed to software: what was made, by whom, and why was written down by people who knew the work.
That is not a complaint about the technology. It is a description of a missing layer. Regulation asks a production to account for itself at the level of the finished work. Our tools mark at the level of a single output. Authorship and rights live at the level of the individual asset. Nothing today carries a claim across all three, so people carry it instead.
None of us can close that gap alone, and a market of incompatible answers helps no one, least of all the artists whose work is at stake. So this paper ends with an invitation rather than a conclusion. We ask broadcasters, studios, producers, vendors, standards bodies and creator organisations to work with us on common standards for ethical AI, production transparency and artistic provenance: a disclosure record that holds good in more than one jurisdiction, provenance that survives post-production, and attribution that reaches the asset and its author, not only the finished master. Build it once, together, and international co-production stays possible. Leave it, and every territory will answer differently, and the people who pay for the difference will be the ones who made the work.
REFERENCES
- List of AI MODELS T&Cs/Policies (?) - to help support ethics / what to keep eyes peeled for with that company / know BEFORE you go! / BEFORE you start to create (?) / what platforms are doing what / changing what (as of Aug/Sept.2026)
- eg: HIGGSFIELD - loads of top creators/producers won’t work with them now, because they treated their creators like crap and did not value their worth
- HOW DO ROLL DOUCHEBAG FREE, when all we want to do is create and make great entertainment, etc?
- Production Identity Whitepaper
- EIDR.org content identity
- EBU - standards?
- AIMICI-
- MOVIELABS etc
- ETC -
- AMD/HP
- GC-SCBest practice (Ben/Aaron)
- Measurement
- Review & continuous improvement
- Why do this - quantifiable benefits, creative friction
- Skills development pathway