It was an example, but what you do in this situation is update the alert deployment process to cater to the failed devices (and it wouldn't likely be as simple as 'these iphones', rather a combination of factors). Focus testing on the specific scenarios/devices that were previously incompatible. There's a point where you're talking about 0.4% of devices, and you'd note that on some risk registry somewhere.

I don't know the acceptable thresholds, but I would expect that if 99% of the population receive the emergency alert, they can give the other 1% a heads-up. Then again, those acceptable thresholds might be 99.9% for all I know.

Either way, sample testing the assortment of phones you have in the lab will only take you so far. At some point you want to trial a full alert to find anything you may have missed.