I think we have 30 years’ experience that clearly shows that dangerous languageswill be misused in critical areas, even if most of us are very careful.
Marcus Ranum once wrote a one-page version of inetd that he thought was secure. He was and is as committed to security as anyone, and had long experience writing software important to the early Internet. Steve Bellovin found a security hole in that one-page program.
I am convinced that a safe language with very tiny holes allowing access to dangerous stuff (like memory management in the kernel) is simply safer. Clearly, we are no where near that right now.
We don't need no stinkin' safety rails, we're smart enough not to walk off that cliff in the first place. And who knows, we may need to walk off that cliff at some point in the future.