Back in 2024, the leading documentary film festival IDFA opened with Piotr Winiewicz’s “About a Hero,” a project designed to test a provocative premise: could an artificial intelligence model successfully conceive and write a film in the distinct style of legendary director Werner Herzog? At the time, the experiment was met with considerable backlash from the documentary community. Filmmakers, including Herzog himself, openly rebuffed the intrusion of artificial intelligence into the traditional craft of cinema, viewing it as an existential threat to human artistic expression and authorship. Like the rapidly evolving technology itself, however, the conversation surrounding artificial intelligence within the film industry has shifted dramatically over the past two years. A growing number of industry professionals and lauded creators are no longer just quietly experimenting with generative tools; they are openly acknowledging their use. Recently, prominent filmmaker Ava DuVernay admitted to incorporating AI technology into her Netflix documentary “14th” to subtly animate public-domain archival photographs, thereby breathing new life into historical periods for modern audiences. Read Also: Liam Neeson Hits the Highway in ‘The Mongoose,’ a High-Octane Fugitive Thriller Directed by Mark Vanselow Rob Savage’s Horror Adaptation ‘Other Mommy’ Delivers a Misfiring, Ineffectual Screen Thrill Rather than shying away from the friction, IDFA tackled the complex issue head-on this year by hosting an industry panel aptly titled “Is AI Stealing My Documentary?” During the session, Wendy Bernfield, CEO of Rights Stuff, delivered a blunt assessment to the audience, explaining that the reality of the situation is far more advanced than many realized. For more than three years, artificial intelligence systems have been systematically scraping and taking digital content before creators were even fully aware of what was happening. Comparing the phenomenon to the early, unregulated days of online video sharing, Bernfield noted that artificial intelligence has closely mirrored that trajectory. For the first few years of widespread deployment, virtually anything available on the open web—including promotional trailers, ad hoc channels, and fast streaming channels—was simply harvested without explicit permission or compensation. The justification frequently offered by developers was that the content was necessary for training foundational models so that automated systems could improve their practical capabilities. Bernfield pointed out that less than two years ago, the technology evolved further into AI aggregation platforms, which combine the output of several distinct AI models into a unified interface for developers and enterprises. Reflecting on her own investigative work into these platforms, she explained that she spent six months interviewing various aggregators and examining the fine print of their contracts. Her inquiry revealed that the core issue was not merely about asking permission and offering financial compensation, but rather about understanding what legal rights, if any, were preserved to protect creators if they chose to license their work. Out of roughly fifteen aggregators currently operating in the market, Bernfield found that only a select few maintain a dedicated focus on harvesting archival image material. While those specific platforms tend to include built-in rights protections for creators, their licensing fees are generally lower than those offered by riskier models that bypass traditional safeguards. Creators, she emphasized, are forced to decide whether they are motivated by short-term financial gains or by a broader commitment to protecting their own legal rights and those of their fellow creators. Elizabeth Klinck, an experienced archive producer at E Klinck Research, echoed these concerns, noting that she is approached almost weekly by filmmakers who have been contacted by artificial intelligence entities inquiring about acquiring their source material. Because the legal and operational landscape remains remarkably new and lacks standardized frameworks, Klinck has increasingly found herself serving as the primary point of contact and unofficial advisor for documentary makers navigating these uncharted waters. Recalling a recent encounter, Klinck described a Canadian filmmaker who discovered a deeply buried opt-out clause designed to prevent her source material from being utilized for AI model training. The contract imposed a restrictive five-day limit on exercising the opt-out option. Ultimately, it required several months of strenuous negotiation to rectify the clause because the filmmaker firmly opposed having her work fed into training pipelines. Klinck realized that countless other creators were in the exact same precarious position, unknowingly signing legal documents with sweeping implications for the future ownership and utilization of their creative output. During the panel, she offered vital advice, urging filmmakers to proactively future-proof their projects against potential legal pitfalls. She warned that if creators incorporate material derived from ambiguous AI processes or unvetted archival sources, the work may fail to hold up under scrutiny from errors and omissions insurance providers or standard copyright clearance protocols. This vulnerability stems from the fact that copyright discussions can shift unpredictably depending on the distribution platform, ranging from traditional theatrical releases to television broadcast and streaming services. Furthermore, third-party rights complicate matters further, as archival material frequently carries embedded rights that demand meticulous re-clearing. When these legal gaps are uncovered during production, Klinck noted that the most practical, albeit time-consuming and expensive, pathway is to negotiate directly with original rights holders. Alternatively, producers can investigate whether specific copyright exceptions—such as historical, creative, or critical approaches—can be successfully argued before legal counsel to justify the material’s use. Ultimately, Klinck stressed the importance of ensuring that any newly acquired archival material comes with explicit, granular rights that legally permit both the use of AI tools and the eventual commercial distribution of the finished project. Esther Van Messel, CEO of First Hand Films, experiences a similar weekly influx of requests for her company’s extensive film catalog. Initially, she admitted, she was deeply impressed by the prospect, envisioning a modern gold rush that could generate significant new revenue streams for her filmmakers. However, Van Messel highlighted a critical demographic limitation in current AI training demands: the vast majority of requests are for bulk volumes exceeding 5,000 hours of footage, which inherently prioritizes mainstream content. Consequently, the data systems training artificial intelligence are predominantly populated by Western, male-centric perspectives and narratives. This very limitation is what makes Van Messel surprisingly open to licensing her catalog to AI developers. She argues that AI models urgently need diverse content, noting humorously that automated systems struggle with everyday domestic tasks because they have never been trained on non-traditional activities. Selling catalog content, she believes, is a way to broaden the intelligence of these systems, provided it occurs within strict, well-defined boundaries. Addressing the shifting market dynamics, Bernfield observed that the buying models utilized by AI developers are transitioning away from massive, indiscriminate bulk acquisitions toward more highly curated, specialized content packages. As mainstream data saturation occurs, niche perspectives suddenly become vastly more valuable. Developers are recognizing significant gaps in their datasets regarding diverse racial backgrounds and LGBTQIA+ representation, prompting a sudden surge of interest in specialized archives. Another major shift involves technical resolution standards. In the early stages of development, AI systems exclusively demanded high-definition 1080p footage, rendering older archival formats largely irrelevant. Once developers secured sufficient high-definition data, however, they began moving backward into older, historic formats, creating a continuous technological curve for rights holders to monitor and navigate. While Van Messel bluntly remarked that the traditional film sales market is currently facing severe stagnation, she emphasized that this downturn presents a silver lining for historical preservation. Because artificial intelligence developers are largely unconcerned with the age of source material, older films with partially cleared rights and decent physical conditions are experiencing a renewed wave of utility. Paradoxically, the intense corporate hunger for archival material is simultaneously threatening the public accessibility of historical resources. Bernfield highlighted that prominent American platforms like Archive.org have begun restricting their free permissions, a trend being mirrored by numerous cultural libraries worldwide. Major journalistic institutions, such as The Guardian, have explicitly opted out of open-access arrangements to prevent their content from being perpetually scraped by automated bots. As a result, the invaluable public repositories that documentary filmmakers have historically relied upon to access exceptional material without prohibitive costs are slowly being eroded. Klinck warned that this trend poses a direct threat to the traditional integrity of archival work. The profession is poised to undergo a profound transformation, shifting away from the traditional discovery of unseen historical material toward roles centered on authentication and policing. She noted that she already encounters job listings on professional networks seeking specialists to authenticate the outputs of new artificial intelligence models at modest hourly rates. As the panel discussion broadened to encompass the ethical dilemmas surrounding the sale of sensitive footage—such as documentary subjects disclosing deeply personal trauma—alongside the massive environmental footprint generated by data-heavy AI operations, Bernfield acknowledged that these questions venture into the broader philosophical debate over whether artificial intelligence is inherently harmful. Concluding her remarks, she emphasized that while the technology presents undeniable ethical challenges, rights holders and producers must focus on what practical measures they can take within their own spheres of influence to minimize exploitation and protect their intellectual property, even when they cannot single-handedly reshape the global landscape. Post navigation Rami Malek Gives Robert Pattinson’s 007 Casting the Ultimate Stamp of Approval: ‘It’s Definitely Rob’s Moment’