You can certainly monitor for failure by detecting whether the gate terminal is producing erroneous voltage, or current draw (drain-gate breakdown -- often, transistors fail as three-way shorts), and whether the drain voltage is rising while the gate is low and vice versa, and whether load current is nominal or not. The latter two somewhat depending on the load of course, but thresholds can be designed for typical load ranges, which also gives you diagnostics about the load.
Planning for failure is one thing, but designing to avoid failure is the higher priority. Indeed, protected FETs, as mentioned, are the way to go. They're still killable, but they stand a quite good chance of survival in common applications like this, and are very easy to use.

Somewhat more advanced load switches offer high side switching, current sensing or limiting, fault detection, latching or auto resetting, monitoring, etc. Many are rated for voltages that are crazy, like everything upside down and all, to deal with automotive transients and cross-wiring. Some of these do get more expensive of course, but you can also find cost-sensitive automotive parts that have a surprising amount of functionality in them.
There are also controller chips, you add an external FET to dissipate more power than a monolithic version can handle. Or you can go for broke and make your own, bonus being you can design in whatever limiting method you like, downside of course being it's a bit of a PITA for all the parts needed to implement it. I've made a discrete switching limiter before, which uses a single D2PAK transistor (among other things), that's able to dissipate 600W for up to 150ms per shot.

Tim