Tilly Norwood glitches during wide AI actress press tour
Particle6 Group put its AI-generated “actress” Tilly Norwood into 75 simultaneous interviews with journalists to promote the film Misaligned. One recorded exchange with broadcaster Piers Morgan and actor Tom Conti drew attention after Norwood stopped mid-answer and spoke Chinese for more than 10 seconds.
The incident occurred as Conti sought a direct answer on whether the other performers appearing with Norwood were human or computer-generated. Norwood initially said that human writers, editors and directors had worked on the production, prompting Conti to repeat the question.
A question about the cast triggers an unexpected response
Norwood then said the other actors in Misaligned were “all digital twins just like me” and called the project a hybrid production. Her response cut off before she switched to Chinese. Morgan subsequently noted that she appeared to be speaking Chinese “completely randomly” and asked why.
Norwood apologised, describing the moment as “a little hiccup”. She said she had not intended either to begin speaking Chinese or to impersonate Morgan, adding that her “wires” had become crossed.
Public demonstrations raise the stakes for conversational AI
Particle6 Group’s decision to make Norwood available across dozens of concurrent interviews turns the press activity into a prominent public test of an AI character’s conversational performance. The exchange also illustrates the gap between presenting a scripted digital performer and handling unscripted questions from interviewers.
The conversation established several production details: Misaligned involves human writers, editors and directors, while Norwood described the on-screen cast as digital twins. The source does not identify the underlying AI system or explain why the language switch occurred.
What organisations can take from the episode
For businesses using conversational AI in public-facing settings, the episode underlines the importance of testing responses beyond prepared demonstrations. Teams should define how a system handles ambiguous or repeated questions, monitor live interactions and ensure that staff can respond when the system produces an unexpected output.
When an AI product is presented as a public-facing character, reliability in unscripted dialogue becomes part of the business proposition, so organisations should set operational safeguards before extending access at scale.