The use of artificial intelligence tools in documentary filmmaking is no longer a future-facing question. AI is already present in research, transcription, translation, classification of archival materials, enhancement of image and sound quality, and, at times, in the creation of images, the reconstruction of events, the generation of synthetic voices, or the protection of participants’ identities. The legal and professional starting point should be balanced: the use of AI in documentary filmmaking is not inherently improper or prohibited. However, precisely because documentary work relies on the viewer’s trust, such use requires heightened caution, transparency, documentation, rights clearance, and a clear distinction between authentic material and synthetic material.
The first distinction is between AI as a working tool and AI as a creator of new documentary material. The use of automated tools to transcribe interviews, classify raw footage, produce initial translations, or search archives does not raise the same level of risk as recreating the voice of a deceased person, turning a still photograph into a video that appears archival, or creating a historical scene that looks as if it were filmed in real time. The closer the use of AI comes to creating evidence, a voice, an image of a person, or an event that appears real, the greater the need to verify, document and disclose.
The guidelines published by the Archival Producers Alliance in September 2024 capture well the perspective of archival and documentary professionals. The document does not call for a ban on GenAI. Rather, it warns against the creation of synthetic material that may appear or sound like an authentic primary source. Genuine archival material was created by human beings at a particular time and place, and carries with it context, ownership, intention and limitations that can be examined and debated. By contrast, material generated by GenAI may appear to carry documentary authority, yet lack a verifiable historical source and clear authorship accountability.
This is the source of the concern about fake archival material. When an AI system creates a realistic image of a historical event, a street, a demonstration, or a specific person, the viewer may understand it as genuine documentation. The risk is not limited to the film itself: such a clip may be extracted, circulated on social media, used in educational materials, or incorporated into other films, thereby blurring the historical record. Even changes that appear merely technical may alter meaning: colorizing black-and-white footage, changing an aspect ratio through generative fill, adding movement to a still photograph, or completing missing details may create the impression of documentation that never existed.
The first legal issue is copyright. Several distinct questions arise: was the AI system trained on protected works; is the output generated by AI protected by copyright, and to what extent; who is entitled to use it; and what do the terms of use of the AI tool provide? In addition, a licence to use an archival clip in a film does not necessarily include permission to upload it into an AI system, use it as reference material, process it, create variations of it, or use it for training. Therefore, whenever AI is used in a material way, it is necessary to examine not only the final output, but also the material fed into the system, the tool’s terms of use, and the scope of existing licences.
The second issue, and often the more sensitive one, is the use of the name, voice, image or likeness of a real person. Documentary filmmaking is about people, and therefore voice cloning, digital replicas and deepfakes are not merely technological effects; they may amount to the use of a person’s identity, voice, reputation, and sometimes rights belonging to a living person or to a deceased person’s estate. In the United States, state legislation is developing in this area, including in California, New York and Tennessee, in order to address unauthorized uses of synthetic voices and likenesses. At the same time, such laws often include important exceptions for news, documentary, historical, biographical, critical or satirical works. The law does not seek to block documentary filmmaking, but it does require an examination of consent, context, commercial purpose, the risk of deception, and whether the viewer may believe that they are seeing or hearing an authentic recording.
Recent examples illustrate this tension. In Roadrunner: A Film About Anthony Bourdain, AI was used to recreate Bourdain’s voice based on sentences he had written but never recorded. The use drew criticism, mainly around questions of consent and disclosure to the audience. By contrast, Welcome to Chechnya presents a different use of similar technology: AI-based face replacement was used to protect the identities of participants at risk. Another example is Endurance, the National Geographic film about Shackleton’s expedition, in which reconstructed voices of Shackleton and members of his crew were incorporated on the basis of historical texts. In all of these cases, the central question is not only whether the text is authentic, but whether the performance, voice or image is presented to the viewer in a way that is not misleading.
Professional bodies and festivals are also beginning to move from a model of silence to a model of disclosure and practical inquiry. In 2025, the Sundance Institute updated its Nonfiction Core Application to include questions about the use of GenAI, including the nature of the use, accountability mechanisms, ethical considerations, and disclosure to audiences and funders. The implication is that the professional market is beginning to expect creators not only to use tools, but also to explain how and why they used them, and how the associated risks were managed.
In Europe, a clearer regulatory framework is also taking shape. Article 50 of the EU AI Act establishes transparency obligations in relation to content generated or manipulated by AI, including the marking, detection and disclosure of deepfakes and certain AI-generated content. The Code of Practice on Transparency of AI-Generated Content is currently voluntary, but its importance lies in the fact that it signals an emerging regulatory and industry standard: providers of AI systems will be expected to offer marking and detection solutions, while users who publish content to the public will be expected, in appropriate cases, to provide clear, distinguishable and accessible disclosure.
Before incorporating AI tools into a documentary work, it is worth pausing to ask a few simple questions: What exactly was created or altered by AI? Is this a technical use, such as transcription, classification or quality enhancement, or the creation of new material that the viewer may understand as authentic? Which tool was used, and what are its terms of use? Were archival materials, voice, images or video used as reference material? Is an additional licence, consent or release required? Was the voice, likeness, name or image of a real person created or recreated? Could the viewer believe that the material is original documentation? Finally, was the use documented and reviewed with legal counsel, a distributor or an Errors & Omissions insurer?
These questions are not intended to deter filmmakers from using AI. They are intended to enable more responsible use. The closer AI comes to creating evidence, a voice, a likeness or an event that appears real, the greater the importance of documentation, rights clearance, obtaining appropriate consents, and providing clear disclosure to the viewer. In documentary filmmaking, transparency does not undermine documentary truth; it is a condition for preserving it.
