The authentication cluster was overloaded, and improperly monitored. Complex systems are like that, never know where the problems will arise. Good clear communication, but it should have been on the AWS blog.