OP-ED - The seatbelt sign is on. Assume the brace position.
Tavus says nearly half the people who video-called its new Griffin model came away believing they'd spoken to a real person, and that it can clone a voice from 10 seconds of audio. It's time to agree a safe word with your family.
I'm not in the brightest mood…
Last Thursday, a San Francisco AI company called Tavus announced Griffin. It's a model that holds live video conversations. It sees you, hears you, nods along, takes interruptions in its stride and generates a face that reacts in real time.
In a live study the company ran itself, 48% of participants came away believing they'd been speaking to a real person. Tavus is holding it back for now, giving access only to select testers, because it says these capabilities make AI harder to tell apart from a person.
The company claims that Griffin can clone a voice from about 10 seconds of audio. That's a lot less than most of the voice notes I send via WhatsApp these days.
I must confess that the growing tendency to anthropomorphise AI is really starting to grate on me. According to the Tavus press release, the company is committed to "teaching machines the art of being human." Ugh.
I mean, do we really want a world where a machine is mistaken for a human? Who's asking for all this 'innovation'? Wait, I guess we are. Awkwardly, I sense that this moment mirrors our capitalist society's preoccupations with at-scale productivity, profit maximisation and convenient value capture.
In April last year, I wrote about the importance of discernment, recounting how actor Keanu Reeves asked a friend's 13-year-old daughter about the difference between real and virtual worlds (within the context of video gaming). "Who cares if it's real?" she apparently quipped. I found that disturbing then, and perhaps equally disturbing now is the determined capitalist intent to build AI tools for a world that has stopped asking.
The study showing Griffin going undetected by nearly half of its conversation partners only tested people who weren't looking for a machine. By Tavus's own account, participants were told they'd be chatting to another participant and were only asked about AI at the end.
There are loads of real-world scenarios where people's guard is similarly down, not least on WhatsApp, where some elderly super-users in my life (who often send me AI-generated inspirational content) spend much of their days casually parsing phones full of forwarded clips and voice notes, and taking video calls from people they love. For the most part, they aren't vigilantly looking out for signs of machine interference in their media and parasocial engagement diets.
Andile Masuku
And I can imagine fraudsters the world over rubbing their hands in gleeful (read greedy) anticipation of Griffin's release. After all, in 2024, an employee in engineering firm Arup's Hong Kong office wired about USD 25 million after a video call where every other face, including the CFO's, was convincingly deepfaked.
In February this year, the chairman of Fideuram, the private bank of Italian banking group Intesa Sanpaolo, received a WhatsApp message that appeared to come from the group's CEO, followed by a call from an AI-cloned voice. As Reuters reported, about USD 108 million was transferred, and roughly a third of that sum is still missing.
In both cases, there was no Griffin-level functionality at play to deepen the deception.
While some might admire Tavus's decision to keep Griffin under lock and key (for now), critics of Big AI's recent waxing lyrical about slowing down roll-outs and doing the responsible thing (because they care) may well point to it as virtue-signalling. Either way, I'm not holding my breath. If Tavus doesn't (eventually) publish the tool, sooner or later, given commercial incentives lucrative enough, some other innovator with fewer scruples than Tavus claims to have will likely publish something similar.
My sense is that the horse has bolted, and we can't reasonably expect AI innovators in Silicon Valley or anywhere else to set the bar for wisdom very high and walk the talk (because they care). So here I am again, playing 'unc', advocating for discernment that starts at home and urging us all to adopt the brace position.
If you haven't already instituted a safe word and a call-back habit with your loved ones, consider this a friendly nudge to do so. Agree the word with your family, ideally in person, around the table and not over WhatsApp. And make it a rule that any request for money, a PIN or urgent help from anyone – however familiar the face or voice – gets a call back on a number you already have saved. You're welcome.
Ugh… What are we doing here, people?
Editorial Note: A version of this opinion editorial was first published by Business Report on 06 October 2025.