One amendment, I'd say it's because existing phones won't let an app have access to listening for a wake word or phrase, and a phone hard codes that to the phone vendor code. Having passive access to microphone and camera and activating and showing what they want to the screen without contending with a platform lock screen that won't play ball with them, that sort of thing. "AI" access wasn't really going to be the challenge.

It's not that they didn't run on existing phones, I could see that, I find it more stupid that they stopped short of just making their device a phone capable of traditional interaction. As it stands it's going to be a subset of capability of phones coming out this year that will likely offer similar "AI" features while also continuing to support traditional hand held usage. If they didn't want to sign up for all that, they probably could have teamed up with someone like Motorola, who might be hungry enough to let Rabbit do their thing on a Moto G variant or something.