Ever since generative AI tools became widely available, they’ve been at the center of controversies. There was the time Stable Diffusion generated images with a warped version of the Getty Images Watermark. Or the time news broke that Apple, Anthropic, Meta and others had trained their AI systems on data sets of film and TV subtitles and pirated books. And of course, most recently, the Ghibli-fication trend that followed ChatGPT’s latest update, encouraged by OpenAI’s own Sam Altman. At the heart of these stories are some fundamental questions around creative rights & ownership—let’s dig into it!

In the creative industries intellectual property (IP) refers to creations that are legally protected. In film and television, IP could include rights to scripts, cinematography, and the final production, among others. Within this, copyright is a key element of IP law, which ensures that creators — or at least copyright owners — maintain control over how their work is used and monetized. Residuals are another crucial component within this framework, providing ongoing compensation to writers, actors, and other creatives when their copyrighted work is broadcast, streamed, derived from, or otherwise exploited beyond its initial release. So what’s at stake with generative AI?

On the input side, the major concern is that there is a huge lack of transparency on the data AI tools have been trained on. This is nothing new, big tech companies are notoriously tight-lipped with their data. Despite little disclosure from the AI companies themselves, reports have uncovered some of the datasets, which unsurprisingly includes swathes of copyrighted material mined from the internet. This data is scraped without permission or consent from the original creator and used to train AI models, again without compensation. This has sparked a wave of lawsuits, including by Getty Images and The New York Times, and protests, for instance by the Writers Guild of America.

As generative AI continues to put a spotlight on IP and Copyright issues, governance—both through legal frameworks and worker-led initiatives—will be essential in shaping fair, transparent, and accountable practices. As a silver lining, we end with a noteworthy exemplar that deviates from the more common practice of data scraping used by tech giants like OpenAI and Anthropic. Moonvalley is a Toronto-based research startup which has created a video generation model called “Marey” which claims to be “clean” AI. According to Moonvalley, Marey was trained only on “ethically sourced” and fully licensed data and creators are compensated for its use. It is too soon to say how effective this alternative model will be or whether it can compete with the tech giants, but developing specialized tools by and for creative workers will be critical to ensuring that generative AI is mobilized to take on sustainable roles and responsibilities in the creative industries.