AI safety and the limits of once-and-for-all fixes

A commenter argues that a single, once-and-for-all solution to AI safety or security is unlikely. They point to nuclear power plant safety, where many solutions for specific reactor designs did not amount to a permanent answer to safety. Their broader concern is human behavior in complex systems: left to themselves under time pressure and competition, people can make poor decisions and become susceptible to groupthink.