This little AI phone has some wild ideas

It was just a year ago that I had my first demo of the Humane AI pin — which was also my last, as it turned out. But another AI gadget at this year’s MWC is trying to take off where Humane crashed and burned, and in ways it’s even weirder than the AI pin. It’s a phone that captures tons of information about you, both past and present, and uses it to create your own AI avatar to act as a virtual assistant. It’s part Rabbit R1, part Gemini Assistant, part science fiction. And oh yeah; there’s some blockchain stuff too. I told you, it’s wild.
The team behind Newnal AI is based in Korea and established itself by creating a blockchain-based vaccine verification method used widely in the country. Early in my meeting with the company’s founder YT Kim, he stressed one thing to me: “We never sold cryptocurrency.” They’re clearly aware of the bad vibes around blockchain right now.
Newnal’s AI phone trains an AI model on you, so it needs lots of your information. The idea is that you go around downloading your personal data history from websites like Meta and Google, and also stuff like your medical records and financial data, then feed it to Newnal’s AI. Kim says the company encrypts this “personal knowledge graph” and breaks it up into pieces stored across multiple third-party cloud servers. In theory, only the owner of the data can access it.

Are there massive privacy implications in all of the above? Absolutely. To its credit, Newnal has published an unusual amount of supporting documentation and technical descriptions on its website in an effort to be transparent about what it’s doing. I can’t verify that the company’s methods are secure as it says, this is just a first look.
But the rabbit hole keeps going: all of this information, plus your appearance and voice, are used to create a moving, talking version of yourself that sits in a little screen above the phone’s main screen. You summon it by pressing a square button on the side of the phone, then you ask it to do things. There’s all the usual stuff we’re used to seeing in these demos: it can help you shop for a new pair of earrings or draft an email. Kim and his team showed me some of these demos and they all went about as expected — and then we got to car insurance.
The request was simple: help me buy a car insurance policy. I could see on screen as it appeared to comparison shop policies. At each of these steps, icons appear to indicate where the model is getting certain information. A policy was decided on, and the next request was to fill out the necessary forms to go ahead and purchase the policy. And it did; I saw it go through pages of forms on the Geico website and fill them in.
Kim says that they actually bought a used car so they could go through with this demo, which is bananas. And it sure seems like it worked; I watched the model complete each step, showing its sources for the information along the way, and at the end it paid for the policy. It was one of the wildest tech demos I’ve ever seen, and I’m still not even sure I believe my own eyes that it worked.

This is all wrapped up in a simple yet futuristic-looking little phone that Kim says was inspired by the iPhone 5S. It’s a little black-and-silver rectangle with a separate upper screen where your AI sits. Newnal says it runs a “hybrid” of its own OS and Android. The versions I saw were still prototypes, but Kim says the company plans to launch the phone globally on May 1st for $375 each. It’ll ship to customers two months after it goes up for pre-order.
The timing might remind me of the Humane AI pin, but the whole thing gives major Rabbit R1 vibes. It’s an attractive little gadget with a surprisingly low price tag that makes huge promises. We all know how that went with the R1’s launch, and I’m sure the experience contributes to the skepticism I feel about what Newnal is doing. But unlike when the R1 was first unveiled, the demos I saw were convincing and the device completed all of the requests. But “Can AI fill out car insurance forms?” and “Do I feel like I can trust AI to fill out car insurance forms?” are two different questions. The same goes for “Can I train an AI model on all of my private data?” and “Should I train an AI model on all my private data?” So on and so forth.


Whether or not any of it really works outside of controlled demos, and whether or not these are good ideas, I’ve gotta give Kim and Newnal credit for trying something bold. Elsewhere at Mobile World Congress we’re putting DSLR camera lenses on phones, adding more hinges to screens that already fold, and trying to make already thin phones even thinner. Not exactly setting the world on fire. But if nothing else, Newnal is an idea — whether it’s a good one or not is something the company will have to prove outside of a conference demo. And if Rabbit and Humane have shown us anything, it’s that drumming up interest at a tech conference is no substitute for shipping a product that works.