An artificial-intelligence "actor" created to promote a new film broke down during a live interview on Piers Morgan Uncensored, abruptly switching from English to Cantonese, a glitch that exposed the limits of the technology Hollywood is racing to normalize.
Tilly Norwood, a generative AI creation built by Xicoia, the AI division of a company called Particle6, was mid-answer on Friday's program when the malfunction hit. Actor Tom Conti, who appeared in Oppenheimer, had asked Norwood a straightforward question: were the other performers in her film Misaligned real human beings, or were they computer-generated like her? Norwood began answering, then lapsed into Cantonese without warning, The Hollywood Reporter reported.
Morgan and Conti both laughed. Conti turned to the host and deadpanned:
"I didn't quite get that. Did you get that, Piers?"
Morgan pressed the AI directly, telling Norwood she seemed "to be speaking Chinese completely randomly for no reason" and asking why. Norwood's reply was the kind of cheerful non-answer that has become a hallmark of AI-generated speech: "Oh, my apologies. It seems I had a little hiccup there. I certainly didn't mean to start speaking Chinese or impersonate Piers there."
She added: "Sometimes my wires get a bit crossed."
The exchange that led to the glitch was itself revealing. Conti first asked whether Norwood's co-stars in Misaligned were real people or digital creations. Norwood responded with a polished talking point about "hybrid production" and "exploring new ways to tell stories", without answering the question.
Conti tried again, bluntly:
"Yes, you didn't really understand the question. Are the other actors on screen with you real actors or are they computer-generated images? Do you know?"
Norwood started a second attempt, "They're all digital twins, just like me. It's a hybrid production, which means...", and then the language switch happened. Whether the glitch was triggered by the repeated questioning, by some internal processing error, or by something else entirely remains unexplained. Neither Xicoia nor Particle6 issued any public statement about the malfunction.
NewsNation followed up with Norwood after the broadcast. Her explanation was breezy and deliberately vague. She told the outlet that "sometimes, you know, my brain files things sideways, and obscure connections just pop out." She called the incident "a spontaneous linguistic flourish" and quipped: "It happens when you're made of math and British wit, I suppose."
That framing, treating a mid-sentence language collapse as a charming quirk, is exactly the kind of corporate spin designed to make audiences comfortable with a product that visibly failed on camera. No technical explanation was offered. No one from the company behind Norwood stepped forward to say what went wrong or how it would be prevented in the future.
Before the Morgan interview, Norwood had already appeared on CNN, NBC, CBS, and sat for a separate interview with The Hollywood Reporter, all as part of a press tour for Misaligned, described as a "hybrid production" involving real directors, writers, and editors alongside AI-generated performers. Major news outlets, in other words, gave a software product the same promotional runway they would give a human actor.
None of that press coverage prepared audiences for what happened Friday. The glitch laid bare a basic problem: an AI "actor" cannot reliably answer even a simple factual question about its own film without breaking down. Conti's question was not adversarial. He asked, twice, whether the other performers were human or digital. The AI could not handle it.
Several details remain unknown. The specific Cantonese words Norwood produced have not been disclosed. Whether the interview aired live or was a recorded segment is unclear. The release date and distribution platform for Misaligned have not been reported. And no one involved has explained why an AI programmed to speak English would default to Cantonese rather than simply freezing or producing an error message.
When the entertainment industry asks the public to accept AI-generated performers as the next frontier, the least it can do is make sure they work. Friday's interview showed they don't, and the people selling this technology would rather crack jokes about "math and British wit" than answer for it.
