Written by Coleton Hast.
Image credit: Language instruction books in Taiwanese aboriginal languages at NCCU by NSaad (WMF)/ Wikimedia Commons, license: CC BY-SA 4.0.
A recent feature article in the Taipei Times, published Sept. 24, 2026, asks in its headline, “Taiwan’s sovereign AI paradox: What exactly do we mean by sovereign?” Despite asking specifically, “Sovereign for whom?,” at no point are Taiwanese Indigenous peoples or their claims to sovereignty mentioned. While frustrating, this absence is not surprising. Headlines that discuss Taiwan’s assertion of sovereignty in the face of Chinese encroachment or US dependence rarely mention the hard truth that Taiwan is a settler-colonial nation with its own Indigenous peoples who assert their own claims of sovereignty. Why is it that questions of sovereignty, AI or otherwise, so often elide the question of Indigenous sovereignty? And how might “sovereign AI” support, rather than undermine, Taiwanese Indigenous sovereignty?
To answer the first question, we might look to the idea of settler-colonial incorporation, which describes the way in which settler societies incorporate their Indigenous populations into the national body and lay claim to the Indigenous inheritance of a given territory. Anthropologist Tomonori Sugimoto, building off the work of Scott L. Morgensen, argues that this “logic of incorporation” exists side by side with the “logic of elimination” in Taiwan. The existence and persistence of Taiwanese Indigenous communities in the face of centuries of settler-colonial occupation is used politically to mark Taiwan’s national, ethnic, and cultural distinctiveness from China. In seeking its own technical sovereignty from primarily Chinese and, to a lesser extent, American AI hegemony, we unfortunately see this same logic of incorporation repeated in the discourse around Taiwan’s sovereign AI development.
While this discourse rarely mentions Taiwanese Indigenous sovereignty, there are several flagship AI projects in Taiwan which incorporate Taiwanese Indigenous linguistic and cultural data. To explore this intersection further, we will examine two such projects: the Trustworthy AI Dialogue Engine (TAIDE) and the Indigenous Languages Research and Development Foundation’s (ILRDF) suite of AI tools.
The TAIDE project is designed to build out Taiwan’s sovereign AI capabilities by developing large language models trained on datasets linguistically and culturally suited for Taiwan. Chang Chau-Iyan, director general of the National Centre for High-Performance Computing, points out that 90% of Chinese-language corpora used in large language model (LLM) training originate in China, with all the linguistic and epistemic problems that entails for Taiwan. While the initial training has focused on a Taiwanese Mandarin dataset using traditional characters, TAIDE and affiliated projects have begun to build datasets of other national languages such as Taiwanese and Hakka, as well as for Taiwanese Indigenous languages. According to Taiwan Panorama, the current sovereign AI corpus for Taiwanese Indigenous languages includes over 61,000 items of Truku voice data and 71,000 items of Tsou voice data, with plans to expand to Puyuma, Kavalan and other languages. While this is a significant corpus, the following project focuses exclusively on Taiwanese Indigenous languages.
To learn more about Taiwanese Indigenous AI development, I recently had a conversation with Omah Canglah, a linguist and fluent Amis/Pangcah speaker who serves as the Amis Product Manager for Ìthuân Khoki Co., Ltd., the company contracted to develop the ILRDF’s AI tools. These tools include freely available automatic speech recognition, speech synthesis, and machine translation for the 17 recognised Taiwanese Indigenous languages and their dialects (excluding the language of the Siraya people, who only gained formal recognition in July of this year). These LLMs are trained on data already held by the ILRDF, and Canglah points out that collaboration with a large, centralised, and Indigenous-led research foundation is necessary to make up the gap in both technical expertise and computational resources. Canglah tells me that beyond their immediate benefit as resources for learning Amis and other supported languages, they help prove to young people especially that their language has a place in the digital world.
Both of these projects are designed to take meaningful steps towards greater representation of Taiwanese Indigenous contexts in Taiwan’s sovereign AI ecosystem. As Pangcah/Amis scholar, Dr. Nikal Kabala’an (Margaret Yun-Pu Tu), points out, LLMs that have not been specifically trained on Taiwanese Indigenous datasets have already been shown to produce disinformation and outright hallucinations. After all, shouldn’t a model intended to reflect Taiwan accurately include the full range of its linguistic and cultural diversity? While avoiding inaccuracies is vital, I would argue that neither AI nor political sovereignty end with the issue of representation. Just as voting representatives in a national body do not make an Indigenous nation sovereign, inclusion of Taiwanese Indigenous data in a model and dataset which they do not control is often antithetical to sovereignty. Indigenous data sovereignty, the right of Indigenous peoples to govern the collection, ownership and application of their own data, has become even more vital in the AI era.
TAIDE, in particular, presents very little opportunity for Taiwanese Indigenous control over their own data. The ILRDF, while state-supported and created by statute, is an Indigenous-led and independent legal entity which owns the copyright to the data which Taiwanese Indigenous language keepers have contributed over the years. While partially dependent on the state, the ILRDF puts meaningful resources and control in the hands of language activists like Omah Canglah. Rather than place these two Taiwanese initiatives into an overly simplistic binary, it is worth contextualising them both within Taiwan and in the wider global struggle for Indigenous data sovereignty. To add to that context, I provide two more examples below of how other Indigenous communities sought to exercise sovereignty with, through and against AI.
Indigenous responses to AI have ranged from acceptance of state-supported initiatives, to local implementations to, yes, outright refusal. The Seminole Nation in Oklahoma on March 7, 2026 became one of the first Native American Tribes to completely ban data centre development on tribal land in a 24-0 unanimous vote. Mekusukey Band Representative Glen Chebon Kernell, who introduced the resolution, said at the time of its passing, “My hope, my prayer, is that the stance we’ve taken will serve as a catalyst for all Indigenous communities to provide resistance to any kind of extractive developments that will harm their people, their communities, and this Earth.” While this is the clearest case of refusal, North American Indigenous Nations have embraced a wide variety of positions towards AI and data centres specifically.
Meanwhile in Aotearoa, the Māori-language platform Papa Reo has increasingly become the prototypical example of what an Indigenous sovereign AI platform can look like. All of Papa Reo’s computational infrastructure is owned and operated by Māori users on Māori land. The AI tools produced by Papa Reo are made up of curated datasets held under the innovative Kaitiakitanga License, which reflects Māori principles of collective data authorship and control. In a blog post reflecting on the creation of Papa Reo, the development team acknowledges the complexities of working with a technology as entangled in colonialism as AI, but ends with a simple final message: “we must respect data.” These examples are meant to demonstrate that Indigenous data sovereignty does not entail a singular outcome, but a process which centres Indigenous self-determination at every step of the way.
To return to my conversation with Omah Canglah, all these cases reveal that we cannot separate AI sovereignty from its political and material context. Moratoriums and local training of specialised models both express and reinforce the data sovereignty of the Seminole and Māori peoples, while reflecting their collective orientations to the world. Sovereign AI in Taiwan should and can look as diverse as Taiwan’s own Indigenous Nations. As Taiwan’s sovereign AI ecosystem grows to include Taiwanese Indigenous languages, it becomes a metonym of the wider treatment of Taiwanese Indigenous sovereignty. I do not know what forms the possible futures of sovereign AI might take in Taiwan, but the least we can do is to make sure that, when asking “Sovereign for whom?”, the answer emphatically includes all Taiwanese Indigenous peoples.
Coleton Hast is a current PhD student in Anthropology at Brown University. His research focuses on the intersection of emergent technologies and Taiwanese Indigenous language revitalization and reclamation. He previously received his Master’s in Ethnic Relations and Cultures from National Dong Hwa University’s College of Indigenous Studies. Coleton can be reached at coleton_hast@brown.edu.
This article was published as part of the special issue on AI Shield and Energy.