Generating exploits with AI and large language models shrinks the time to target software flaws, giving security teams scant time to patch. Can enterprises adapt?