PETALING JAYA: Malaysia’s two locally developed artificial intelligence (AI) initiatives – ILMU (Intelek Luhur Malaysia Untukmu or Excellent Malaysian Intelligence for You) for sovereign capability and NurAI for syariah-aligned guidance – position the country to lead Asean in trustworthy, values-based AI, said AI risks and safety adviser Chan Tzu Kit.
He said ILMU – the country’s first homegrown large language model (LLM) launched by YTL AI Labs Sdn Bhd – strengthens sovereign capability with a locally tuned multimodal model for government, education and SMEs, while Zetrix AI Bhd’s NurAI – the world’s first syariah-aligned, Islamic values-based LLM – addresses a gap to curb misinformation and expand services in Islamic finance, halal and public guidance.
“Sovereign, locally trained multimodal models also mean sensitive Malaysian data doesn’t have to leave the country,” he told SunBiz.
Chan described the move as a step-change for Malaysia, shifting the nation from being an AI user to an AI maker.
However, he stressed that trust will hinge on safety and governance.
Chan called for both initiatives to publish model cards and undergo independent evaluations, particularly on Bahasa Melayu performance, refusal rates and factuality in sensitive areas such as religion, health and finance.
He also urged authenticated sources and human-in-the-loop for sensitive queries, local red-teaming with bug bounties to test for jailbreaks and data leaks, and the use of content watermarking for official outputs.
Chan further proposed a single incident-reporting portal with annual independent audits, and on-premises or virtual private cloud deployments with strict access controls for the public sector.
“Malaysia can be the place where AI is both powerful and principled. Guardrails turn good demos into dependable systems.
“If NAIO (National AI Office) makes ILMU and NurAI early pilots for national and Asean guardrails, Malaysia can lead the region pairing language and values fit with transparent evaluations and clear accountability,” he said.
Several countries in the region are also developing their own AI models. Singapore has launched SEA-LION, a family of open-source LLMs for Southeast Asian languages, and MERaLiON-AudioLLM for speech/text. Indonesia is building Sahabat-AI, focused on Bahasa Indonesia and local languages. India has multiple projects, Bharat GPT, BharatGen, AI4Bharat and Sarvam AI, all aimed at multilingual Indic capabilities. Thailand created the Typhoon LLM for Thai, while South Korea’s Upstage Solar Pro 2 is a frontier model rivaling GPT-4.
In a separate matter, Chan highlighted other immediate threats Malaysia faces from unchecked AI development – the potential loss of up to half of Malaysia’s palm oil exports by 2027 – as AI-driven labs develop cheaper synthetic alternatives, and the displacement of up to one in three Malaysian workers, with AI replacing entire categories of jobs faster than the country can adapt.
Universiti Tun Hussein Onn Malaysia head of Environmental Management and Conservation Research Unit (eNCORe) Dr Noraini Ruslan said if synthetic palm oil becomes commercially viable too soon, oversupply could drive down crude palm oil prices and severely impact the income of smallholders, repeating the history of Malaysia’s natural rubber industry.
“The decline of rubber, once the country’s dominant commodity, came swiftly after synthetic rubber flooded global markets and displaced natural rubber in many applications. Malaysia cannot afford to let palm oil follow the same fate,” she told SunBiz.
Regarding this matter, Chan advocated for setting up a dedicated institute to conduct full-time research on AI safety and economic transition strategies.
He stressed that Malaysia needs only RM1 million per year to kickstart such an institute, one-tenth the annual budget of the National Innovation Organisation. “This institute would focus on anticipating economic disruptions and formulating policies before mass layoffs and export shocks occur.”
Chan said Malaysia has a track record of being “three years too late” on critical digital issues: it delayed enacting the Personal Data Protection Act, cybersecurity laws and digital watermarking standards.
“Delay with AI safety could prove catastrophic as Malaysia can be ‘pelanduk mati di tengah-tengah’ (mousedeer crushed between two elephants) caught between the US and China’s AI arms race.”