Enterprises connecting large language models to email inboxes and calendar apps face a new, documented threat: researchers have shown that an attacker can plant hidden instructions inside URL previews ...