Indeed, I would say that capability-safety is the next plateau after memory-safety, but most languages are only barely glimpsing the capability horizon.
This is personally one of my major pet peeves, because I like to write secure code, and I think you very rapidly end up half-reconstructing capabilities work if you do, and it is not hard to notice that you basically get no help whatsoever from languages. It is still clearly a nearly-invisible problem to the vast majority of developers. You need something almost Haskell class simply to assert "Please, at least provide me some evidence that you ran some sort of permission check." to access a resource.
For "you ran some sort of permission check" I'm a great fan of the React.js `dangerouslySetInnerHTML` interface, which accepts a corresponding `{__html: ...}` object that has at least nominally been properly vetted.