A report finds humans acting as safety checkpoints on AI coding agents fail to catch roughly a third of dangerous requests, like exposing credentials.
Continue to AI University →