Sovereign AI

AI sovereignty is an extension of national responsibility for education, citizens’ data and cultural heritage. Countries need the capacity to decide what AI systems learn, how they are used and who is accountable. Today, about 75% of the computing power of the world’s 500 largest AI supercomputers sits in one country and 15% in another, and most countries lack the expertise to assess frontier models (Independent International Scientific Panel on AI, 2026).

Sovereignty does not mean isolation: no country owns the whole AI supply chain. A practical approach is a hybrid one, where non-sensitive tasks can use external cloud services while learner records, assessments and administrative data stay on nationally governed systems under domestic law. Drawing on several AI suppliers also reduces exposure to price changes, export controls and outages.

Sovereignty is also critical for language and culture. Some open models now work across 140 languages, yet models trained mainly elsewhere may poorly reflect national curricula and local languages, and exclude Indigenous Knowledge. Open models can be further trained on authorised local material, with communities deciding what is used and how knowledge is represented (Lewis, 2020). Japan’s experience shows that such adaptation can improve performance in a national language without building models from scratch (Fujii et al., 2024).

In February 2026, the India Student Community for AI launched within the GPAI community, and over 60 students joined its AI Impact Summit satellite event. With support from COL and its Commonwealth Educational Media Centre for Asia, and GPAI expert centres in Canada, France and Japan, it has become the Commonwealth Student Community for AI Governance, with cohorts in more than 20 countries by August 2026.

The community works through four pillars: national student coalitions, interdisciplinary research, evidence-based policy advocacy and institutional reform, from curricula to academic standards.

Language and data sovereignty run through every cohort’s agenda. In Sri Lanka, students working with the Ministry of Education prioritise Sinhala and Tamil language technology with human oversight. Kenyan students are developing Kiswahili voice AI and, in August 2026, submitted a memorandum on the draft national AI policy calling for a youth seat on the National AI Council, subsidised computing for academic research and a right to explanation for automated decisions.

In Pakistan, students at Allama Iqbal Open University are working on tiered AI policy and rural AI labs; Maldivian students focus on Dhivehi; Australian students on safeguards for Indigenous cultural and intellectual property; and African and Caribbean cohorts on open-source tools in local languages, shared climate-prediction tools and cultural data sovereignty.

Their message is consistent: open-source language technology can counter algorithmic bias and the neglect of local priorities. These views can inform the Commonwealth Youth Forum, whose outcomes will be presented at CHOGM in St John’s.

Regional policy is moving the same way. In the Caribbean, a regional AI task force has called for a harmonised, rights-based framework, a Caribbean data commons and resilient regional computing.

University partnerships supported by COL, including through the Gaborone to New Delhi Compact (COL, 2026), can pool computing capacity, expertise and methods, so smaller countries can take part while retaining control of their data and cultural resources.