Just last week, eighteen major Japanese media companies formally requested that OpenAI cease using their content to train its Sora 2 video generation model. The companies, represented by the Content Overseas Distribution Association (CODA), asserted that a significant portion of Sora 2’s outputs closely resemble copyrighted Japanese content, indicating potential unauthorised use of their intellectual property as training data. The Japanese government supported this position, with officials emphasising the importance of protecting the country’s cultural assets within AI development.
This moment exemplifies the fact that over the past 24 months, the world’s leading media brands have confronted a new category of risk. One that exposes systemic gaps in how they secure their most valuable assets: their data, IP and licensing.
This risk has been accelerating dramatically because the explosion of generative AI models spanning text, image, music, and video has collided with a wave of enterprise adoption across media and entertainment. In the race to capture AI’s creative and operational advantages, speed-to-market has been key. And while AI has undeniably driven unprecedented transformation across the industry – enabling personalised content experiences, automating production workflows, and unlocking new creative capabilities – rapid capability growth on all fronts has amplified potential exposure of core IP and proprietary data and underscored the need for governance to be embedded from the ground up.
This was a topic that dominated the conversation at Advertising Week New York, a recent gathering of the biggest brands at the forefront of the media landscape. In meetings with senior media executives across the city, Ula Nairne, Stelia’s VP of Media and Entertainment highlighted a shift in tone. “The central question was no longer whether or how to adopt AI, but rather how organisations can scale AI capabilities securely while maintaining control of what makes them competitive.”
Why media executives are rethinking how their data, rights, and models are governed
The proliferation of large language models has created a new intersection of intellectual property, data licensing, and AI that traditional security frameworks weren’t originally designed to address.
This began to come to light in December 2023 when the New York Times sued OpenAI and Microsoft, claiming the companies had built their AI models by copying and using millions of the publication’s articles without permission, content that took decades and significant investment to create.
The New York Times argued that these systems now directly compete with its content as a source of information. Cases like these exposed a fundamental gap in IP security. Existing licensing frameworks had no provisions for AI training, and legal frameworks designed for human infringers weren’t equipped to address automated systems that could process millions of works simultaneously. Critically, scraping at scale had occurred before organisations were able to recognise the need for technical safeguards against this category of risk.
Shortly after, major music labels including Sony Music, Universal Music Group and Warner Records filed suit against two AI music startups Suno and Udio for copyright infringement described as occurring on “an almost unimaginable scale”. Here the challenge extended beyond unauthorised training; AI-generated outputs were infringing on licensed IP, with no clear party to hold accountable when a model produced music that replicated protected works.
These cases revealed the initial collision point, and represent a moment of realisation for media companies. As Ula, our VP Media & Entertainment, put it, “AI systems suddenly demonstrated they could access, ingest, and even replicate some of the industry’s most valuable assets, proprietary content that took decades and significant investment to create”. Importantly, these early cases all involved external actors operating without permission, resulting in responses through legal action, stricter access controls, and updated terms of service.
For example, publishers such as the BBC, Reuters, and The Guardian restricted crawler access to protect content from scraping. Others, including Getty Images, updated licensing terms to explicitly prohibit AI use, laying early groundwork for more structured data partnerships.
Turning this moment into an opportunity, Getty launched its own licensed AI model in partnership with Nvidia, based on fully authorised content. Generative AI by iStock is a text-to-image platform specifically designed to make stock photos.
What’s next?
More companies are beginning to explore similar initiatives, fine-tuning and developing their own models.
As AI innovation continues to accelerate and raise both new opportunities, as well as potential exposure points, what is certain is that the ability to architect systems, policies and technology stacks that protect owned data and models is the new competitive advantage. This involves moving beyond responsive action to proactive architectural thinking. Treating IP protection as a post-deployment compliance exercise is no longer viable, by then the exposure has already occurred. Instead, it must be foundational to how AI systems are designed, deployed, and governed.
Equally, while governments debate AI-specific IP frameworks – from the EU AI Act to Japan’s proposed AI transparency standards – regulatory clarity is not keeping pace with technological reality.
In the coming weeks, we’ll examine some of the architectural considerations that enable AI transformation at scale while embedding security as the foundation for competitive control.
