A digital avatar of a head of government is no longer science fiction, but an experiment that could soon become a daily reality in Malaysia. Prime Minister Anwar Ibrahim – the country's tenth leader – is preparing to launch PMX AI, an artificial version of himself designed to interact with citizens and, according to Bloomberg, even to handle payment links. The rollout, expected within days, marks a turning point in how public institutions use artificial intelligence.

The move is bold: a political leader delegates part of the public dialogue to an autonomous system, bypassing traditional communication channels. PMX AI, whose name nods to the prime minister's position as the tenth officeholder (PMX stands for Prime Minister X), is not a simple chatbot, but an entity meant to act proactively, managing operations ranging from information assistance to financial transactions. For those working with on-premise deployment and data governance, the central question is the underlying infrastructure.

No technical details have been released yet, but the context suggests a Large Language Model (LLM) that could be self-hosted on local stacks to ensure sovereignty and control. For a government, the choice between cloud and on-premise infrastructure is not trivial: it involves Total Cost of Ownership, compliance with national regulations, and the need to shield citizens' data from external access. From an AI-RADAR perspective, an on-premise approach might offer advantages in terms of audit and security, but it poses hardware challenges – consider the VRAM required to run large models or managing a real-time inference pipeline.

Using an avatar with transactional capabilities inevitably raises questions about algorithmic transparency. What happens if PMX AI makes a mistake in a payment or provides incorrect information? Responsibility falls on the political leader, but the technical tool must be governed by clear rules. In Europe, initiatives like GAIA-X advocate for digital sovereignty; similar approaches could inspire Asian countries, especially when an AI represents the head of the executive branch.

The Malaysian trial arrives at a moment when governments are experimenting with generative AI tools to reduce bureaucratic load and improve responsiveness to citizens. PMX AI is not an internal assistant, but a public interface that duplicates – at least in part – the prime minister's presence. For organizations evaluating on-premise deployment, the reference architecture might include frameworks such as vLLM or TGI, with quantized models to balance resource consumption and latency, but all of this remains speculative. What is certain is that Malaysia is drawing a new line: if the initiative succeeds, the demand for local infrastructure capable of hosting sensitive workloads is bound to multiply.