top of page
Search

Intelligence Piracy: The Actor Who Was Never There

  • Writer: Oliver Walsh
    Oliver Walsh
  • 11 minutes ago
  • 6 min read

By Oliver Walsh, Head of Evidence Services, Rouse




In the autumn of 2025, a new actress began doing the rounds. She had a name, Tilly Norwood, a headshot, a showreel and, before long, talent agencies reportedly taking an interest. She had never drawn breath. Tilly was generated entirely by software, and when the actors' union SAG-AFTRA responded, its objection doubled as a definition of the problem. She was, it said, "a character generated by a computer program that was trained on the work of countless professional performers."


She is not a person. So whose craft is she made of?


That question is the whole of this article. In the first piece in this series, I argued that artificial intelligence has opened a third generation of piracy, one that steals not a finished work but the capability behind it, and that music was the first industry to feel it. Film and television are next, and the exposure is arguably greater, because a screen performance carries something a recording does not: a face.


What was actually stolen?


No film was pirated to make Tilly Norwood. No single performance was copied. Yet anyone in the industry understood at once that something belonging to real actors had been taken. Not a movie. The accumulated craft of thousands of performers, distilled into a synthetic who competes with all of them and pays none of them.

Film and television now have two distinct assets to defend. The first is the individual performer's identity: their face, voice, movement and presence. The second is the studio's library and visual style, the decades of finished work that a model can learn from. Both are under the same pressure, and both are being lost in the same way.


Every generation of piracy, on screen


The pattern is familiar from music. The first generation was content piracy: camcorder recordings, DVD rips, torrented films. The second was platform piracy: illegal streaming and IPTV services, organised release-group ecosystems, counterfeit storefronts. In each case the target was the film, or the pipe that delivered it.


The third generation does not want the film at all. It wants the performer and the studio's style. Deepfakes place real actors in scenes they never shot. Generative video models produce finished footage from a text prompt. Synthetic performers like Tilly are cast in real productions. Dead actors are revived and living ones de-aged. The asset under attack has moved from the film to the platform to the identity and craft within them.


The library is a teacher model


The mechanism is the one I described for music. A studio's decades of footage, and an actor's entire body of work, are training data. A generative video model learns faces, voices, mannerisms and house style, then reproduces them on demand. No single film is copied. The capability to recreate a performer or a look is extracted.


This is no longer theoretical. In February 2026, SAG-AFTRA condemned ByteDance's Seedance video model as "blatant infringement," pointing to unauthorised recreations of well-known performers, including Tom Cruise and Brad Pitt. Whatever the eventual legal finding, the objection is the third generation stated plainly: the model had learned the actors well enough to reproduce them, and no one had asked.


A synthetic performer is not no one. It is everyone it was trained on.

When the theft has a face


Because a performance carries identity, the core legal asset in film and television is the right of publicity, the control a person has over their own likeness and voice. That makes the harm more personal than in almost any other sector, and the disputes more pointed.


In May 2026, the actress Q'orianka Kilcher filed a claim against James Cameron and The Walt Disney Company, alleging that her face, captured when she was fourteen, was used as the structural foundation for a computer-generated character. The allegations are contested and unproven, and the case will take years. But the question it raises is exactly the one the industry now has to answer: where does inspiration end and the appropriation of a real person's biometric identity begin? Scarlett Johansson's continuing disputes over deepfakes and a synthetic voice are pushing the same question into policy.


There is a legitimate version of all this, which is worth stating. When James Earl Jones agreed that his Darth Vader voice could be recreated from archive recordings, he did so knowingly and was paid. Consent and compensation are possible. Intelligence piracy is simply what happens when they are absent.


The response, and its limits


The industry is not standing still. SAG-AFTRA's 2023 strike was, at its heart, a fight about AI and digital replicas, and the TV and theatrical agreement its members ratified in June 2026 strengthened those protections further. Around it sits a growing patchwork of law: state rights of publicity, the proposed federal NO FAKES Act aimed at unauthorised digital replicas, the ELVIS Act protecting voice, and the TAKE IT DOWN Act addressing non-consensual deepfakes.


The limits matter as much as the protections. A performer's identity can be protected, but a studio's visual style largely cannot, and the boundary between training that is transformative and training that is theft remains unsettled. That uncertainty is likely to persist for years, which means the outcome of individual disputes will turn less on settled law than on what can actually be proven.


Why this is an investigator's problem


And proof is the difficulty. Like the music cases, audio/visual intelligence piracy often leaves no copied artefact to hold up in court. The evidence lives in three questions.


  • Was a model trained on this library or this performer, which is a matter of data provenance and dataset forensics?

  • Is a given synthetic output derived from a specific person, which calls for facial and voice biometric comparison and output forensics?

  • And who built and monetised the system, which is the familiar climb through infrastructure, financial and human attribution that my colleagues run every day, and which the Intelligence Attribution Pyramid sets out.


Studios, agencies and performers will increasingly need litigation-grade evidence to enforce likeness rights, support claims and negotiate licences from a position of strength. That evidence has to be identified and preserved early, often before anyone is sure a dispute exists. Detection and legal drafting are not enough on their own. Attribution, the discipline of establishing who did what, is the piece most often missing, and it is investigative work.


The face in the frame


The first generation of screen piracy copied the film. The second copied the platform. The third recreates the performer, and it can do so without touching a single frame of the original.


Tilly Norwood is often described as an actor who never existed. That is only half true. She was assembled from the work of people who very much did, and who were never asked. The next article turns to sport, where the contest itself is the content, but the real prize has quietly become the data.


This is the second in a six-part series on intelligence piracy, the third generation of digital theft, from Rouse's Evidence Services team.


References:
Rouse logo RGB-03.jpg

©2024 Rouse. This site is produced for the purposes of information only.

  • alt.text.label.LinkedIn
bottom of page