In "The Fall and Rise of Formal Methods", Peter Amey gives a pretty good description of how I expect things to play out w.r.t. Friendly AI research:

Good ideas sometimes come before their time. They may be too novel for their merit to be recognised. They may be seen to threaten some party’s self interest. They may be seen as simply too hard to adopt. These premature good ideas are often swept into corners and, the world, breathing a sigh of relief, gets on with whatever it was up to before they came along. Fortunately not all good ideas wither. Some are kept alive by enthusiasts, who seize every opportunity to show that they really are good ideas. In some cases the world eventually catches up and the original premature good idea, honed by its period of isolation, bursts forth as the new normality (sometimes with its original critics claiming it was all their idea in the first place!).

Formal methods (and I’ll outline in more detail what I mean by ‘formal methods’ shortly) are a classic example of early oppression followed by later resurgence. They arrived on the scene at a time when developers were preoccupied with trying to squeeze complex functionality into hardware with too little memory and too slow processors.

...[But now] formal methods... are on the rise. And why not? What is the alternative? If we don’t use them, then our tool box contains just one spanner labelled ‘test’. We know that for anything that justifies the label ‘critical’, testing will never be enough...

For the curious, Amey also wrote a nice overview of successes and failures in formal methods.