AI Affairs, home

Saturday 3 October 2026

Government

AI Alliance reports Project Tapestry training milestone and wider Asia ties

The alliance says two trials kept training data at participating sites. Vietnam is discussing a sovereign model, while BharatGen helped run a trial across India and Australia.

A hand holds a smartphone displaying an AI app beside an artificial intelligence book.
Photo: Sanket Mishra via Pexels

The AI Alliance said on 30 September that Project Tapestry had completed its first technical milestone, using two trials to train models across four geographically distributed sites while keeping participating organisations’ data local. In an announcement distributed by PR Newswire, the alliance also described expanded sovereign AI collaboration with Vietnam and India during United Nations General Assembly Week in New York.

Key points

  • The AI Alliance says two proofs of concept coordinated model training across four sites without moving the underlying training data from participating sites.
  • Alliance leaders met Vietnamese officials in New York to discuss a sovereign AI model aligned with the country’s priorities.
  • BharatGen ran one trial with Monash University across India and Australia, according to the alliance.
  • BharatGen plans to host a Project Tapestry workshop in Mumbai on 15 October.

Four sites test consortium training

Project Tapestry is the AI Alliance’s effort to build a consortium for developing foundation models and sovereign AI systems. Its proposed arrangement lets participating institutions contribute data, computing capacity, expertise and model improvements while retaining control of their underlying data. The alliance says the two proofs of concept tested whether organisations working across different countries and infrastructure could train models together under that arrangement.

For the completed milestone, the alliance said, training was coordinated across four geographically distributed sites. It described the work as a demonstration of building blocks for collaborative training and adaptation, rather than the completion of a national sovereign AI system. The trials addressed a practical condition for institutions that want to work on a shared model without placing all their training data under one organisation’s control.

The teams also modified a large language model with the aim of improving its understanding of social and linguistic aspects of Indian and Vietnamese cultures, according to the alliance. That work sits alongside the consortium-training trials in Project Tapestry’s first milestone. The alliance says it intends to expand work on cultural alignment and data governance as the project proceeds.

Anthony Annunziata, the AI Alliance’s co-founder and board chair, called the trials a move from architecture to working technology. His account was narrower than a claim that a shared sovereign model is ready for use: organisations in different countries had trained models together while retaining control of their own data, he said. Project Tapestry’s plans extend from that technical work to collaboration with national institutions.

Vietnam discusses a sovereign AI model

AI Alliance and Project Tapestry leaders met Vietnam’s General Secretary and senior government ministers in New York during UN General Assembly Week, according to the alliance. Christopher Nguyen, Project Tapestry’s chief architect, identified General Secretary and President Tô Lâm as taking part in the meeting. The alliance said the parties signalled that they wanted to work together on a sovereign AI model shaped by Vietnam’s AI policy goals.

Those priorities include Vietnamese language, culture, knowledge and national uses, the alliance said. Vietnam has participated in Project Tapestry since its early stages, according to the announcement. The expanded discussions concern how a country might develop a model for its own purposes while drawing on the work of institutions elsewhere in the consortium.

Ambassador Hoang Anh Tuan, Vietnam’s consul general in San Francisco, described the project as an opportunity to turn international collaboration into national capability. In the alliance’s announcement, he said that meant developing AI able to understand Vietnamese language, culture, knowledge and priorities, alongside engineers and institutions able to operate and improve the technology.

BharatGen trains across India and Australia

India’s consul general Binaya Srikanta Pradhan hosted Project Tapestry leaders at the Consulate General of India in New York during UN General Assembly Week, the alliance said. Ganesh Ramakrishnan of BharatGen, Ritwik Banerjee of Stony Brook University and Kaushik Bhatta of the AI Alliance briefed him on the consortium’s approach. The briefing also covered Stony Brook’s addition to the project, according to the announcement.

The alliance describes BharatGen as an Indian government-backed foundation model stack with publicly accessible source code, and said it conducted one of the two trials overseen by Maneesh Singh, its vice-president of machine learning. It trained jointly with Monash University across India and Australia, the alliance said, with each institution’s data remaining in its home country. The account identifies the partners and locations for one trial, while the alliance describes the wider milestone as spanning four sites.

Pradhan linked India’s participation to the country’s linguistic and cultural diversity. “AI built for its languages and cultures preserves agency,” he said in the alliance’s announcement. BharatGen’s role in the trial places an Indian institution within Project Tapestry’s technical work as well as its discussions with diplomatic representatives.

The alliance also said the Gates Foundation named Ramakrishnan a 2026 Goalkeepers Champion at Goalkeepers New York for his leadership of BharatGen. Project Tapestry contributors presented a paper on federated approaches to AI sovereignty at a workshop held alongside ACM HCOMP 2026 in Washington, DC, according to the announcement.

BharatGen plans to host a Project Tapestry workshop in Mumbai on 15 October, bringing technical and institutional leaders together to plan the next phases of India’s participation, according to the AI Alliance.

Topics: Public sector, Sovereign AI