The year 2027 marks the mid-point of the International Decade of Indigenous Languages and a moment in the new era of generative artificial intelligence (GenAI) when we might expect multilingual large language models (LLMs) to enhance the future prospects of Australia’s Aboriginal languages. The presumed opportunity is for people to reclaim their ancestral languages with the support of chatbots that have been trained on all available content. However, LLMs perform poorly on Indigenous languages, not just because of the dearth of training data, but because oral cultures are out of view of LLMs trained on monolingual text collections. What, then, does cultural survival look like in the GenAI era? We seek nuanced answers to this question that fall between the extremes of blind acceptance and outright rejection. Key themes include: (1) relationality, where knowledge is embedded in the land and transmitted at the right time and right place to the right people, not the free-for-all assumed by GenAI; (2) hybridisation, where communicative practices may be Indigenous-western hybrids, not the pure forms found in documentary records; and (3) embodiment, where language serves local participation, and is not the information-seeking activity commonly assumed by designers of GenAI. This presentation will draw on experiences arising in collaborations with First Nations speech communities and language centres in the north and west of Australia, and shared with permission.