AI Actress Tilly Norwood Went Viral After Speaking Chinese—But Her Video Chat Service Raises Privacy Questions
AI actress Tilly Norwood has gone viral after an unexpected glitch during an interview with Piers Morgan caused her to start speaking Chinese. The clip had been viewed more than 8 million times by last night.
Norwood has become the subject of widespread controversy and mainstream media attention as the lead in an upcoming AI-generated film.
Her creator, Eline van der Velden, runs Talking Tilly, a service that allows people to video-call AI characters. I tried the service myself and examined the fine print that many callers may overlook.
Talking Tilly scans your face before the call starts
An automatic age check must be completed before a call can begin.
Your selfie video is analyzed by Didit, a Spain-based identity-verification provider, to estimate your age. If the system cannot make a clear estimate, you can upload a government-issued photo ID instead.
(Bleeping Computer)
Xicoia Ltd., the British company behind Tilly, says selfies are sent directly from your device to Didit. The company says no facial-recognition data or biometric templates are created and that selfies and ID images are not retained after the check. Instead, it maintains approximate age ranges and reference numbers.
The age check cannot be skipped and applies to callers worldwide. It was added to the service’s terms of service this month, alongside a workplace ban and new automated safety systems.
The AI service also analyzes your voice and emotions
Facial scans are not the only form of analysis performed during a call.
Talking Tilly monitors camera footage and listens to the caller’s tone of voice to infer their emotional state. According to the company, this allows the AI character to respond in a way that matches the caller’s mood.
The privacy policy states that this monitoring “cannot be turned off for individual calls.”
Both the age verification and mood analysis rely on legitimate interest rather than consent as their legal basis. Xicoia’s version history shows that this selection was made in September.
Calls are recorded, transcribed and processed live by a US-based provider. Character responses are generated using Google’s Gemini model through the conversational video platform Tavus.
Automated safety filters can stop calls by mistake
The service also uses an automatic classifier to screen call transcripts for policy violations, including profanity. If a call is flagged, recording is put on hold.
One of the three calls I made—a conversation about the weather and news headlines—was stopped for “hate and abusive language” that never occurred.
The service’s policy says human reviewers can examine recordings that are flagged in error. In either case, the recordings are permanently deleted after 24 hours.
Talking Tilly offers five free minutes before asking users to pay
The first five minutes are free. After that, callers are asked to purchase additional time. Prices start at £0.99 for a single five-minute session, £13 for 15 minutes or £22 for 30 minutes. Each person can purchase a maximum of 35 minutes.
(Bleeping Computer)
The fine print is particularly important because Talking Tilly is a limited-time service.
All minutes, whether free or paid, will expire when Talking Tilly permanently shuts down on September 27. Any unused minutes will be forfeited.
Transcripts are retained for up to eight weeks and may be reviewed by Xicoia employees or third-party partners. The AI characters can also remember previous conversations to personalize future interactions, although users can request that those memories be deleted.
Why Talking Tilly requires facial age verification
Facial scanning to verify that someone is over 18 before they can talk to a chatbot may seem unusual, but it reflects the UK’s broader direction on online age verification.
Adult websites serving visitors in the UK were required to use identification or facial age estimation from July 2025 under the Online Safety Act. The government’s social media ban for under-16s is also expected to require similar checks for anyone opening a new social media account from spring 2027.
The government’s announcement particularly alarmed AI companies. Despite documentation about the service’s safety measures, Tilly’s safety page claims that “Tilly is not one of us.”
As a result, regulatory compliance decisions in the UK mean callers are being scanned regardless of where they are located, from Manchester to Ohio.
Was Tilly Norwood’s Chinese-speaking glitch an accident or marketing?
Xicoia, founded by Tilly creator Eline van der Velden, describes the character as an awareness project designed to demonstrate how far AI video technology has progressed.
Exclusive: AI actress Tilly Norwood glitches during interview with Piers Morgan and real-life actor Tom Conti and starts speaking Chinese…
Watch the full interview at 7pm BST.
https://t.co/1nTb79BH8D @piersmorgan | @TillyNorwood pic.twitter.com/DDkveWvVzz
— Piers Morgan Uncensored September 18, 2026
When I asked Tilly about the Piers Morgan moment through the service, she said it “wasn’t exactly an approved answer.”
Despite being presented as nominally British, Tilly also described the weather using Fahrenheit.
Talking Tilly is scheduled to go offline permanently on September 27 at 11:59 p.m. PT, just days after the Piers Morgan appearance.
Was the glitch simply a coincidence, as Tilly cheerfully claimed, or was it a well-timed stunt? For now, only the people behind the AI character know for sure.
Join Mikko Hypponen and security leaders from the NFL, Chanel and Atlassian for a two-hour digital summit about AI-speed attacks, what defenders should stop doing, and how to verify, decide, fix and revalidate at machine speed.
Source: www.bleepingcomputer.com