In the summer of 2023, archival producers Stephanie Jenkins, Rachel Antell, and Jennifer Petrucelli found themselves at the forefront of a new frontier in filmmaking, when an unexpected encounter with “fake archival” sparked the formation of the Archival Producers Alliance (APA).

About one year later, on September 13, 2024, the APA released its “Best Practices for Use of Generative AI in Documentaries,” aiming to support filmmakers navigating the increasing integration of artificial intelligence in nonfiction storytelling. The guidelines has since been endorsed by dozens of documentary and filmmaking organizations, including Video Consortium. APA co-founders Jenkins and Antell shared what led them to advocate for transparency of AI use in documentary filmmaking.

"A little over a year ago, we started to see generative AI come into our work on a film we were working on, in the form of what we call 'fake archival,'" recalled Rachel Antell, an archival producer with over two decades of experience in major TV and streaming documentaries. The issue with this discovery was twofold: AI-generated visuals were not only entering editing rooms but also doing so without any established guidelines for transparency—a cornerstone of nonfiction work. "It was right when ChatGPT was coming online, and we realized this was going to hit the documentary world,” said Antell.

The group's first action was to pen an open letter signed by over 100 industry professionals, highlighting growing unease over "fake archival" materials created by GenAI. Feedback to the letter was “overwhelmingly positive," said Antell. "It became really clear that people were very hungry for guidance."

AI technology has progressed quickly in recent years. A pivotal moment came in 2021 with the documentary "Roadrunner: A Film About Anthony Bourdain," which stirred controversy by using an AI-generated version of Bourdain's voice. Fast forward to today, and AI tools have become widespread, affordable, and remarkably sophisticated. These technologies can now not only mimic voices with uncanny accuracy but also create stunningly lifelike visuals. It takes just a simple text prompt and a few clicks to generate something that could potentially be taken for reality in a documentary.

For eight weeks, about 15 volunteers from across the U.S. and U.K. met weekly, discussing documentary ethics, the nature of truth and technology's role in storytelling. Stephanie Jenkins, with more than 15 years of experience as an archival producer, said that the guidelines were formed from those weekly discussions. "It became like my favorite graduate school seminar that I've never attended," she said.

The group grappled with complex questions: What's the difference between generating a photograph and licensing a real one? What's the difference between human bias and algorithmic bias? Most crucially, what impact will generative AI have on the historic record and how we perceive the past? While AI offers new possibilities, it also presents significant tradeoffs for documentary filmmakers.

The guidelines emphasize transparency and the "irreplaceable value of primary sources" — video and audio recorded by humans — and encourages filmmakers to avoid altering source material or to alert audiences when they have done so. But the guidelines don’t forbid AI altogether: “We were pretty quickly swayed to the fact that it absolutely does have a place in documentary,” said Antell. The guidelines suggest naming AI tools and companies in the credits.

Some positive applications of generative AI include 'Welcome to Chechnya' and 'Another Body' – where generative AI technology is used to protect the identities of interview subjects, with full consent and transparency toward the audience.

'Another Body,' directed by Sophie Compton and Reuben Hamlyn, chronicles a college student's journey as she confronts the unauthorized use of her likeness in deepfake pornography.

The stated goals of the guidelines are to reaffirm the journalistic values present in documentaries and preserve the fact-based nature of documentary, both susceptible to erosion when technology is misused or undisclosed. A case in point is Netflix's April 2024 true crime documentary "What Jennifer Did," which displayed signs of AI-editing such as blurry boundaries and misshapen hands. Eventually, a producer confirmed that archival background alterations were made for anonymity purposes.

"There aren't really any bad actors, certainly not within the documentary world," Antell said. "Documentary is always short of money and time. AI can really address both those issues. So I understand when filmmakers are struggling, why there's a real instinct to just run and grab it."

At its core, the APA's initiative aims to safeguard one of the fundamental principles of documentary filmmaking: trust. Antell underscores the critical nature of this trust, stating, "Once the trust between filmmaker and audience is broken and people don't think of documentaries as something that really are representing truth, it's very hard to win back people's trust."

In an era where AI can blur the lines between fact and fiction, these guidelines serve as a crucial compass for maintaining the integrity of nonfiction storytelling. By fostering transparency and ethical use of AI, the APA is working to ensure that documentaries continue to be a trusted medium for conveying truth and reality to audiences worldwide.

To read the full guidelines, click here.

Thanks to Hyacinth Empinado for contributing to this piece.

Alex Clark is a video journalist, documentary filmmaker, and adjunct associate professor at Columbia University's Graduate School of Journalism. His work includes hosting the Emmy-nominated series "Glad You Asked" for Vox, producing the PBS NOVA film "Crypto Decoded," and editing for the Peabody-nominated film "The Picture Taker.”

Rough Cut Magazine is edited by Monica Gokey.

Rough Cut Magazine is VC's digital mag for and by industry thought leaders, doc filmmakers, and video journalists across the world.