Liability works on the principle that "if it's good enough for Google, it's good enough for me." A bank cannot realistically vet every vendor, so they rely on the OS maker to do the heavy lifting.

Even if they wanted to trust a third-party OS, they would need to review them on a case-by-case basis. A hobbyist OS compiled by a random volunteer would almost certainly be rejected.

Google doesn't provide an API or data set to figure out what the current security patch level is for any particular device. Officially, OEMs can now be 4 months out-of-date, and user updates lag behind that.

Your guess is good, but misses the point. Banks are worried about a couple things with mobile clients: credential stealing and application spoofing. As a consequence, the banks want to ensure that the thing connecting to their client API is an unmodified first-party application. The only way to accomplish this with any sort of confidence is to use hardware attestation, which requires a secure chain-of-trust from the hardware TEE/TPM, to the bootloader, to the system OS, and finally to your application.

So you need a way for security people working for banks to feel confident that it's the bank's code which is operating on the user's behalf to do things like transfer money. They care less about exploits for unsupported devices, and it's inconvenient to users if they can't make payments from their five-year-old device.

And this is why Web Environment Integrity and friends should never be allowed to exist, because Android is the perfect cautionary tale of what banks will do with trusted-computing features: which is, the laziest possible thing that technically works, and keeps their support phone lines open.