I'd love to see what, if any, evidence they have to back up the claims they're making.
These old comical soundboards of celebrity voices are imperfect as the audio often has different volume levels and subtle background noise, but illustrates the general principle. With an intense enough effort, you can record enough audio from a person to put together a natural conversation. It takes some effort, but is not outside of the technical capabilities for even a smart 14 year old to set this up.
https://www.101soundboards.com/boards/10716-arnold-schwarzen...
Unless these financial security systems have a way to look for very subtle, unnatural gaps in the audio or the consistency of the audio quality, the voice pattern that exists could be a perfect match with this kind of tactic.
The handful of times I've had to contact Fidelity, their customer service has been exceptionally professional. The only annoying part of the process is trying to insert a long password via the phone dialpad to verify my identity. I've been given the suggestion to setup their MyVoice feature, but have resisted setting that up because it seems like there's a possibility it could be bypassed.
And even ignoring that voice ID is insecure, I still think for the amount of complexity going into voice ID, there are much simpler systems that would be just as easy to use. The difficulty of getting someone to enroll their voice into a database can't possibly be lower than the difficulty of getting them to read a code off the back of their card, or to set up a passcode, or to read off an OTP from an app.
Maybe balance checks are trivial enough where someone doesn't care if they get hacked, but it's still weird because there should be a secure login method over the phone that gets used for everything. That there are separate login methods for different operations is weird and probably not good security. Why bisect it?
If the normal login methods are too cumbersome or hard, well... that's another situation that's worth questioning. There are easy login methods for services that are much more secure, why isn't the normal phone authentication using those? Why is it using a method that's so cumbersome that they need a different login method to check bank balance?