Skip to content

Security1 publisher2 min readPublished

Xicoia face-scans every Tilly Norwood caller worldwide before the first call connects

Xicoia made a biometric age estimate mandatory for every Talking Tilly caller this month and cites legitimate interests as the legal basis. The service closes on September 27; transcripts can be held for eight weeks.

The Watch · Security desk

Illustration accompanying Xicoia face-scans every Tilly Norwood caller worldwide before the first call connects

What happened

  • Talking Tilly routes every first-time caller through a video selfie that Spain-based Didit analyses to estimate age, with a government photo ID upload as the fallback when the estimate is unclear.
  • The check cannot be skipped and applies to callers anywhere in the world, and it entered the service's terms only this month, alongside a workplace-use ban and new automated safety systems.
  • Every call is also analysed for mood, with the system watching the caller's camera feed and listening to their tone of voice so the character can respond in a way that fits it.
  • Talking Tilly shuts down permanently on September 27, and every minute a caller bought or was given free expires with it.

Compiled by The WatchSomething wrong?How this is made

Why it matters

  • exposure A rule aimed at UK visitors now reaches everyone: a caller in Ohio hands a selfie to a Spanish processor and a live camera and voice feed to US ones, and the only way to decline is to skip the call.
  • decision Teams shipping consumer-facing AI video now have a worked template to copy, and copying the legitimate-interests basis means defending that decision without a consent record to point at.
  • contradiction Xicoia's safety docs say "Tilly is not a companion" while the UK's under-16 ban announcement named AI companion chatbots for 18+ enforcement, and that label decides who has to gate.
  • precedent Once the under-16 checks land in spring 2027, third-party age-estimation vendors sit inside the data path of consumer products that never handled identity documents before.

A caller's video selfie goes from the device to Didit in Spain, which returns an age estimate, and an unclear estimate triggers a request for a government photo ID [1]. The call that follows is recorded, transcribed and processed live by US providers, and Google's Gemini generates the character's replies through the conversational video platform Tavus [7]. Xicoia says the selfie travels straight to Didit, that no faceprint or biometric template is created, that neither the selfie nor any ID image survives the check, and that it keeps an approximate age band and a reference number instead [6]. Those are the company's statements about its own processing.

Xicoia runs all of this without asking the caller to agree. It gives legitimate interests, not consent, as the legal basis for both the age check and the in-call mood inference, and its own version history dates that decision to September [5]. The privacy policy says the mood analysis "cannot be switched off for an individual call" [4].

The automated safety layer misfires. A classifier reads each call transcript for abusive language and withholds the caller's recording when it flags one [8]. The BleepingComputer reporter who tested the service made three calls, and one of them, a conversation about the weather and news headlines, was withheld for "hateful or abusive language" that never occurred [9]. A human reviewer can release a wrongly flagged recording, and every recording is permanently deleted after 24 hours regardless [10]. The deletion clock gives that reviewer 24 hours to act [6].

The pricing runs backwards. After the free five minutes, 99p buys a one-time five-minute starter, £13 buys 15 minutes and £22 buys 30, capped at 35 purchased minutes per person [11]. That is 19.8p a minute for the starter against 86.7p for the 15-minute pack and 73.3p for the 30, so the bulk packs cost between 3.7 and 4.4 times the starter rate per minute [1][2]. Every minute expires on September 27 and unused minutes are forfeited [12].

The source describes no intrusion and no exploited flaw [7]. Xicoia added a mandatory worldwide biometric check to a consumer product mid-run, with about eight days of service life left when the reporter tested it [3]. Recordings last 24 hours and transcripts up to eight weeks, 56 times longer, and Xicoia staff and third-party partners may read them [5][13]. A transcript from a call on the final day can sit until roughly November 22 [4]. Xicoia's founder Eline van der Velden describes the character as an awareness project intended to show how far AI video has come [16].

What to watch

  • Whether any UK regulator tests the legitimate-interests basis for the age check and the mood inference before the service closes on September 27.
  • What happens to stored transcripts and the character's conversation memory after shutdown, given retention runs up to eight weeks.
  • Whether the spring 2027 under-16 rules treat AI characters like Tilly as companion chatbots for 18+ enforcement.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories