The challenges of multilingual communication in an emergency are immense. Providing accurate, trusted information at speed is difficult in English. When local communities speak multiple languages, and have varying levels of proficiency in English language, those difficulties can seem impossible to resolve. In this presentation, we share findings from our research investigating the challenges of multilingual communication in an emergency. Qualitative interviews were conducted with people from refugee and migrant backgrounds living in regional Australia. Their experiences and insights highlight the challenges of accessing good quality information in an emergency that is necessary to ensure their safety. Placing these experiences in the context of international case studies and research, we identify the factors that are key to effective, timely and trusted communication in an emergency. We then consider the extent to which AI tools are able to meet these requirements. Aiming to separate the reality from the hype, we conclude by offering suggestions for inclusive and effective uses of AI in emergency communications, which might reduce language barriers without increasing inequality of access to essential information.