The Monthly Roundup: Angela Suriyasenee on Southeast Asia’s sovereign-AI race
ASPI researcher Angela Suriyasenee explains why sovereign LLMs matter for security—plus her must-reads.
This is a special edition of ASPI's Daily Cyber & Tech Digest, a newsletter that focuses on the topics we work on, including cybersecurity, critical technologies, foreign interference & disinformation. Sign up here.
Follow us on Bluesky, LinkedIn, or X.
Welcome to another edition of The Daily Cyber & Tech Digest Monthly Update! Each month, an ASPI expert shares their top news picks and provides their own take on one key story. This time, Angela Suriyasenee, Researcher — Cyber, Technology & Security Program, shares her perspective.
Southeast Asia’s sovereign-AI race intensifies as Thailand readies ThaiLLM
The AI race in Asia isn’t just about innovation, it’s also about sovereignty and security. From Delhi, to Bangkok and Seoul, governments are building their own large language models (LLMs) to reduce reliance on foreign platforms, secure sensitive data and defend against disinformation. This drive is already taking shape in the region. Singapore has introduced SEA-LION, a government-backed open-source LLM, while India is advancing Bhashini, a state-led translation platform. Japan’s NTT has developed tsuzumi, and in South Korea, Naver Cloud operates a hyperscale Korean-centric LLM open to the public.
Thailand is the latest to step in, committing 25 billion baht (AUD$1.18 billion) for AI development over 2026-2027, and plans to launch ThaiLLM in September 2025.
Thailand’s ruling coalition is under strain, with an early election is possible before 2027. The recent border crisis with Cambodia has also fueled disinformation. These pressures will test not only Thailand’s politics but also its ability to defend its information space, driving Bangkok to develop its own LLM as part of a broader Indo-Pacific push for sovereign AI.
Unlike larger Indo-Pacific countries that focus on a single flagship LLM, Thailand is pursuing a multimodal approach. Pathumma, built by the National Electronics and Computer Technology Center, is a multimodal model that processes Thai text, speech and visuals. It is already being used to summarise parliamentary debates and digitise government records. The Big Data Institute’s open-source ThaiLLM also aims to support government, academia and industry, while the National Science and Technology Development Agency is training sector-specific models for tourism, healthcare and transportation in partnership with Siam AI Corporation. This decentralised approach reflects a simple insight: sovereignty in AI offers more resilience against disruption when it rests on a diverse ecosystem, not just one model.
By training these systems on Thai language, slang and culture, Thailand isn’t just improving user experience, it’s building national security tools that can capture nuances that global systems, primarily trained in English, might overlook. These varied models give an edge in spotting disinformation, countering election meddling, coordinating during crises, and stress-testing cyber defences. Singapore’s SEA-LION and India’s Bhashini highlight how localised models can also sharpen cyber defence by analysing open-source intelligence and flagging region-specific threats.
Thailand has recognised that building trust in AI tools is as crucial as technical capability. ThaiLLM offers open access but with controlled usage. Pathumma runs through secure Application Programming Interfaces (APIs), which are essentially gateways for data exchange, to safeguard information. Policymakers are also linking these projects with ASEAN’s AI governance principles, emphasising transparency, accountability and ethical use. However, safeguards on paper don’t eliminate risks. Weak API security could still leak sensitive data, biased training sets might skew outputs, and foreign-supplied components could introduce hidden backdoors. Just as importantly, the same models designed to detect false narratives can also be weaponised to generate and amplify them at scale.
Thailand’s challenge lies not only in building these models, but in keeping them credible, secure and sustainable. Japan’s tsuzumi model can be a reference point for compact, lightweight and energy-efficient design for long-term operational sustainability. On the flip side, South Korea’s HyperCLOVA X illustrates how scaling up language models can reinforce strategic autonomy by reducing dependence on foreign platforms. Thailand, with limited resources, will need to weigh which approach best balances sovereignty, resilience and cost.
Bangkok’s push for sovereign AI is more than a national project — it’s a test case. Success could secure digital autonomy for smaller economies and support regional progress in AI. Stronger digital sovereignty and capabilities in Bangkok would help fortify Indo-Pacific resilience, improving the region’s ability to withstand disinformation, reduce reliance on external powers and build a more stable and self-reliant ASEAN.
My must-reads
Carnegie Endowment for International Peace
In this report, Elina Noor and Binya Kanitro examine how Southeast Asian developers are challenging the dominance of Western-led language technologies by designing LLMs tailored for regional languages and contexts. With over 1,200 languages spoken across the region, new models like SEA-LION, SeaLLM, and Sailor represent deliberate efforts to address the “low resource language” gap and embed local perspectives in AI. The paper underscores that, beyond technical achievement, these models are a statement of agency and identity, reshaping how Southeast Asia participates in the increasingly global, yet still Western-centric, AI ecosystem.
Fulcrum
In this piece, Surachanee Sriyai highlights the real-world impacts of mis- and disinformation campaigns, and risks that come with mobilising public support in the cyber domain especially during times of conflict. Sriyai examines the regulatory gap in the intentional spreading of misleading information in armed conflict, and how public participation in digital discourse can be a force multiplier for states’ objectives that can lead to the incitement of nationalism and physical violence.
ASEAN
This guide on generative AI governance highlights risks such as disinformation, deepfakes, and biased outputs, and the importance of accountability, data security, and content provenance for national LLM initiatives like ThaiLLM. It outlines how countries can mitigate these risks while promoting digital sovereignty and public trust, and discusses the importance of regional coordination to strengthen resilience against information operations and support responsible AI innovation across Southeast Asia.
For more on China's pressure campaign against Taiwan—including military threats, interference and cyber warfare, check out ASPI’s State of the Strait Weekly Digest.
