Failure detection is via alarms to trigger patient action based on the continuous glucose monitor (which has a different set of reliability issues) as well as patient symptoms.
Hypoglycemia becomes symptomatic long before blood sugar is low enough to result in death or serious debilitation and T1D patients know their symptoms well. The risks are not nearly as dramatic as you're suggesting as one isn't/shouldn't be relying on glucagon to prevent severe hypoglycemia, I don't think any system is designed or being conceived to operate in such conditions.
Hypoglycemia isn't really much of a problem anymore with current CGMs and pumps.