Tilly Norwood was built to talk. During an interview with Piers Morgan, she abruptly started speaking Cantonese.
The AI-generated performer switched from English to Cantonese for more than 10 seconds during a recorded interview published Sept. 18, 2026. Norwood returned to English after Morgan questioned the change and called it a “little hiccup.” Particle6, the studio behind her, later said the episode was not a malfunction.
Particle6 also made Norwood available for 75 simultaneous journalist interviews during the press tour, according to TechCrunch. At that scale, operators need to know what an AI system said, why it behaved unexpectedly, and how quickly a human can intervene.
What happened during the interview
Actor Tom Conti asked whether Norwood’s co-stars in Particle6’s in-development film Misaligned were human or computer-generated. Norwood first answered by describing the human writers, editors, and directors involved before Conti clarified his question.
Norwood then said the other performers were digital twins like her, stopped mid-sentence, and began speaking Cantonese. Morgan asked why she had switched languages, and Norwood returned to English and apologized.
Particle6 publicist Michelle Waldron told Forbes that the switch was “not a malfunction” and suggested Norwood may have interpreted something she heard as Cantonese. Waldron also said Talking Tilly, the version used for interviews, is separate from the acting version and runs on different tools.
Particle6 founder and CEO Eline van der Velden later suggested Norwood may simply have been demonstrating her language abilities. No public technical account has established what triggered the switch.
Separate CNN reporting on Norwood’s production process showed Particle6 refining prompts during an acting exercise, but that demonstration does not establish whether anyone intervened during the Morgan interview.
The control problem behind AI avatars
Norwood’s interview is not the only example of AI being pushed into direct conversations with people. OpenAI is testing brand-sponsored conversational agents that businesses can shape around their websites and services, putting AI directly in front of customers.
Reliability remains a practical concern. Google says its new Gemini 3.8 Live models can switch among 97 supported languages, while its model documentation warns that the systems can hallucinate and may experience slowness or timeouts.
Organizations deploying similar systems should know who can change prompts or guardrails, which records are retained, and who can stop a session when something goes wrong. South Korea’s proposed security rules for autonomous AI agents similarly call for audit logs, shutdown mechanisms, access controls, and human approval for high-risk actions.
In September 2025, SAG-AFTRA criticized Norwood as a synthetic performer and said it opposed replacing human performers with synthetics. For enterprise deployments, an autonomous avatar speaking on an organization’s behalf needs auditable records, escalation procedures, and a reliable human override.
Particle6 has not publicly released a technical explanation for Norwood’s language switch. Until one exists, the episode shows that conversational fluency does not guarantee predictable behavior.
Let us teach you How to Talk to AI for free! Try our six-minute course at The Neuron Academy and learn a few simple ways to write better prompts and get more useful results from AI, or browse our other AI course for free for seven days. Check out all the lessons here →
Read the full article here
