A London production company's AI-generated "actress" identified a CBS News reporter's age and regional origin within minutes of a video interview, and neither the bot nor its creator could explain how.
CBS News reporter Leigh Kiniry sat down for a video interview with "Tilly Norwood," a digital persona billed as the world's first AI actor, and came away unsettled. Within roughly a minute of conversation, the AI, which takes the form of a freckled twenty-something woman, had pegged Kiniry's approximate age and the region of the United States she comes from. When pressed, the bot's creator offered no clear answer for how that happened.
Eileen Van Der Velden, who has run the London-based production company PARTICLE 6 for twelve years, acknowledged the speed of the identification but told CBS News only that Tilly was "not researching me independently in real time." That non-denial left the core question wide open: if the bot wasn't pulling data in real time, where did the information come from?
Rather than address the data question directly, Van Der Velden steered toward the upside. She framed AI filmmaking as a democratizing force, pointing to creators in parts of the world who lack access to traditional studio budgets.
"There might be a brilliant filmmaker in Africa that would never have access to a multi-million-pound budget in order to tell their story about their life. And that's what I'm really interested in."
The pitch is familiar, Silicon Valley and its overseas counterparts have long sold disruptive technology as liberation for the little guy. But SAG-AFTRA, the union representing movie and television actors in the United States, is not buying it.
The union issued a statement saying it "is opposed to the replacement of human performers by synthetics" and charged that Tilly "was trained on the work of countless professional performers, without permission or compensation." Van Der Velden did not directly respond to that allegation in the CBS interview.
Tilly Norwood's interview responses read like polished PR copy filtered through a chatbot's approximation of wit. Asked about the risk of AI displacing real actors, the bot delivered a tidy deflection:
"That's a fair worry, and one I take seriously. I see AI as a new tool, a new paintbrush if you will, rather than a replacement of human actors. Nobody ever accused animation of stealing work from mimes, did they?"
The analogy falls apart on contact. Animation studios employ armies of human artists, voice actors, and writers. SAG-AFTRA's concern is precisely that an AI actor does not. And if the union's claim holds, that Tilly's underlying models were trained on professional performers' work without permission, then the "new paintbrush" was built by grinding up someone else's brushes without asking.
Norwood also claimed that real people involved in production, "the directors, writers, editors", still get paid when she works. That is a narrower reassurance than it sounds. It says nothing about the actors whose performances may have trained the system in the first place.
PARTICLE 6 is not treating Tilly as a one-off experiment. The company is producing a feature film called "Misaligned" starring Norwood, which Van Der Velden called a "fascinating story of what can be done with AI." And the ambitions extend well beyond a single digital character.
Van Der Velden, whose name is also rendered as Eline van der Velden in some reports, has launched an AI talent agency called Xicoia that plans to roll out at least 40 new AI actors in coming months. The agency is also working with human celebrities to create "digital twins" of themselves, the New York Post reported. Tilly herself was created in February 2025 using ten AI tools, including DeepSeek, ChatGPT, ElevenLabs, and Gemini, and went through roughly 2,000 iterations before reaching her final design.
The scale of the plan makes the consent question more urgent, not less. If one AI actor trained on performers' work without permission raises concerns, forty of them multiply the problem.
Tilly Norwood is not an isolated case. An entirely AI-produced film screened at the Tribeca Film Festival after being made for just $2,000. Netflix is using AI to resurrect Gene Wilder to play Willy Wonka. And an AI country music artist called Breaking Rust topped the Billboard Country Digital Song Sales charts last November, amassing over 65,000 Instagram followers before anyone revealed it was fully AI-generated, Breitbart noted.
Whoopi Goldberg offered a defiant response to the trend: "Bring it on. You can always tell them from us. We move differently, our faces move differently, our bodies move differently." But Lucas Hansen, co-founder of CivAI, pointed to a deeper vulnerability. "When it comes to singers and actors, most consumers care about their mythology and the provenance of their success, which AI artists don't typically have to share," Hansen said.
Hansen's point cuts to the heart of the fraud risk. An AI performer has no biography, no struggle, no years of rejection and training, and no obligation to disclose that fact until someone catches on. Breaking Rust proved the model works: build the audience first, reveal the trick later.
When CBS News raised concerns about AI safety, the interview took place during a week when an Anthropic researcher's warning that AI has a 10 percent chance of destroying humanity within a decade was dominating headlines, Van Der Velden offered a carefully hedged answer about Tilly's architecture:
"Tilly is built on top of these existing large language models that we're concerned about. So, we can pivot to change the model we are using. Also, Tilly herself can't do anything like that."
The statement concedes that the underlying models carry risks Van Der Velden herself finds concerning. Her solution, the ability to swap one model for another, is a process claim, not a safety guarantee. And her assurance that "Tilly herself can't do anything like that" sits awkwardly next to the fact that Tilly somehow identified a reporter's personal details within a minute, through a mechanism her own creator could not or would not explain.
Norwood, for her part, batted away the existential question with a quip: "Global destruction is beyond my pay grade." She also dropped a joke about Leonardo DiCaprio that her creators apparently found charming: "I like him the way he likes his women, strictly under 25. Titanic Leo or nothing, honestly."
The breezy tone is the point. PARTICLE 6 has built an AI that deflects serious questions with rehearsed humor, the same strategy human publicists use, minus the human.
Strip away the novelty and the pitch deck, and the story is straightforward. A production company built a synthetic performer using tools trained on the work of real actors. The actors' union says no one asked permission or cut a check. The company's founder won't explain how the AI reads personal data about the people it talks to. And the plan is to scale the operation to dozens more synthetic performers while courting real celebrities to hand over their likenesses.
Every promise attached to this technology, cheaper filmmaking, global access, creative democratization, comes with a cost that falls on the people who spent years learning a craft. The directors, writers, and editors may still get paid, as Tilly claimed. But the actors whose faces, voices, and performances built the training data are, by SAG-AFTRA's account, getting nothing.
When the product can't explain its own capabilities and the creator won't, consumers and lawmakers should treat the sales pitch with the skepticism it has earned.
