Two visions of justice arrived on screens in 2026. In the American screenlife thriller Mercy, a detective accused of murdering his wife wakes strapped to a chair before an AI judge on a giant screen.
He has 90 minutes to prove his innocence. As the AI retrieves digital traces of his life, it continually recalculates his probability of guilt. Archive, investigation, trial, and sentencing collapse into a single screen-centered apparatus.
In China, The Out-of-Control “Box,” a short film from Beijing’s Haidian District Procuratorate based on an actual case of kaihe (“opening the box,” a form of doxxing that digs up and exposes people’s private information), reenacts a prosecutorial investigation into the online exposure of virtual-streamer voice performers and people around them. A “prosecutor + data examiner” team connects scattered digital traces on a wall-sized display.
Both imagine a trial by screening: traces are retrieved, sorted, and connected until a case and a person’s guilt take shape on screen.
But evidence does not reveal itself. Someone has to make it.
A photograph, message, testimony, video, or data point must be captured, preserved, selected, connected, authenticated, and made legible as proof. To say that evidence is made is not to say that it is made up. It is to recognize human labor—archival as well as technological—through which traces come to count as evidence.
The burden of proof is also a burden of media practice.
That burden is vividly staged in She Said—Jodi Kantor and Megan Twohey’s 2019 account of their 2017 investigation into Harvey Weinstein, adapted into a film in 2022, which follows the painstaking work of piecing together testimonial voices, documents, interviews, and corroborating traces into a publicly credible case.
Feminist scholarship has long traced the gap between speaking about sexual violence and being believed. Legal scholar Deborah Tuerkheimer in her 2017 work describes a “credibility discount”confronting women who report sexual violence. In their 2023 book, Sarah Banet-Weiser and Kathryn Claire Higgins show how sexual-violence claims circulate through a mediated “economy of believability.”
But who must do the work of making violence and its aftermath evidentiary in the first place?
For survivors, that work can be archival, forensic, logistical, and affective: preserving messages and screenshots, documenting injuries, reconstructing timelines, authenticating images, navigating reporting systems, and repeatedly returning to painful material—as the experiences of Shiori Itō in Black Box Diaries and Xianzi in her case against Zhu Jun make strikingly visible.
A 2026 Harvard study of seven major online platforms found that survivors reporting nonconsensual intimate images still encounter opaque and inconsistent procedures that leave much of the burden of reporting harm on them. These are forms of media labor: the work required to make an experience legible as evidence.
My own research of the history of socialist media offers an unexpected way into this contemporary crisis, not because practices of public accusation in socialist China and AI-generated image-making are the same, or because one led to the other. They did not. But this history makes visible something easy to overlook now: evidence has to take form.
Long before digital databases, making tiezheng (“irrefutable evidence”) in Chinese socialist campaigns involved assembling documents, photographs, names, objects, accusations, and stories into what I call an incriminating catalogue.
Known variously as zui’eshi, heishi, or choushi—histories of crimes, “black histories,” or shameful histories, these catalogues turned fragments and traces of a life into an incriminating whole through accumulation and arrangement.
But an incriminating catalogue can capture more people than the person it was built to judge. Years ago, while researching case files from socialist China, I encountered nude photographs of a woman in a dossier assembled to incriminate a man.
Not simply an object in his case, she was one of two women who had accused him. Her testimony helped make the case against him, yet the same evidentiary apparatus made her visible, too—exposed to judgment and shame. She incriminated and was incriminated.
I will not identify the archive, the case file, or any of the people involved here. That withholding is intentional. Evidence can be indispensable to justice, yet the person making evidence can herself become evidence.
Generative AI twists this bind further. Phones, watches, cars, and homes already produce what Andrew Guthrie Ferguson calls “self-surveillance”: everyday traces that can become incriminating clues against ordinary people. Generative AI introduces the inverse possibility—an incriminating-looking trace of an event that never occurred.
Governments are responding. In the United States, the TAKE IT DOWN Act requires covered platforms to provide a process for removing qualifying nonconsensual intimate images, including AI-generated ones.
China requires explicit and metadata-based identifiers of certain AI-generated and synthetic content. These measures operate in very different legal and political contexts, but both confront a basic problem: what happens when an image can no longer be assumed to be a trace of an event?
However, the answer cannot simply be fewer traces. A message, photograph, recording, or timestamp may be precisely what makes violence evidentiary and accountability possible. Women, in particular, can now bear a double burden: confronting persuasive images of sexual acts that never occurred while still struggling to make violence that did happen believable.
Evidence is necessary. But evidence is neither self-evident nor innocent. Evidence has multiple lives. Its traces are “ephemeral remains”: fragile enough to disappear, yet stubbornly recyclable—preserved, copied, recontextualized, and made to speak again.
Today, there is an abundance of fabricated visibility alongside an unequal burden of making real violence evidentiary. That burden should not fall so heavily on those already harmed.
The task, then, is not to abandon evidence but to redistribute the labor of making it. Legal, media, research, technological, and care institutions must bear a greater share of the sustained archival, forensic, and logistical work of documenting violence and its aftermath, establishing their evidentiary force, and limiting the unwanted recirculation of intimate traces.
Justice must make violence evidentiary without requiring those who experienced it to make themselves endlessly available as evidence.
About the Author: Belinda Qian He is Assistant Professor of East Asian Cinema and Media Studies at the University of Illinois Urbana-Champaign and a Public Voices Fellow through The OpEd Project. She is working on her forthcoming book, Expose and Punish: Cinema and Incriminating Media in an Age of Endless Reckoning.
