A Black Hat demonstration revealed how maliciously crafted inputs can hijack AI assistants to perform unauthorized actions.
