There is a date that quietly invalidated much of the roadmap, and it is not in the document anywhere: 30 November 2022, the day ChatGPT launched. Malaysia’s AI-Rmap was researched and written in 2020–2021, in a world of classical machine learning, analytics and IoT. The generative-AI wave that broke a year later changed what “AI policy” even meant — and the roadmap had no answer for it.
This is the most forgivable of the roadmap’s problems — no one saw the speed of it coming — but also the most consequential, because it dated the plan’s entire conceptual frame, not just its targets.
The AI the roadmap was written for — and the AI that arrived
The 2021 frame
- Task-specific machine learning & analytics
- AI as a tool deployed in defined “use cases”
- Data and IoT as the strategic frontier
- Governance focused on bias, privacy, transparency
- Capability built project-by-project via consortia
The post-2022 reality
- General-purpose foundation models & generative AI
- AI as a horizontal layer across every task at once
- Compute & access to frontier models as the frontier
- New risks: hallucination, deepfakes, model misuse, IP
- Capability bought via hyperscaler cloud & partnerships
How the gap showed up in practice
By 2024, generative AI was not a theoretical concern in Malaysia — it was mainstream. At the launch of the national AI ethics guidelines, the Digital Minister cited a Microsoft–LinkedIn finding that 84% of Malaysian knowledge workers were already using generative AI, above the global average. A plan that barely mentioned the category was, by then, describing a world that no longer existed.
The government’s own responses acknowledged the shift. The voluntary responsible-AI principles were said to be under growing pressure from generative AI; the MOSTI minister noted that existing laws — the Communications and Multimedia Act 1998, the Penal Code — would need amending to keep pace, and that it would take time for guidelines to mature into law. Each statement is, implicitly, an admission that the 2021 settlement had been overtaken.
Why this is structural, not cosmetic
A roadmap can absorb a missed target. It cannot easily absorb a change in the nature of the thing it governs. Generative AI did not just add new items to the to-do list — it changed what infrastructure mattered (compute over bespoke data pipelines), what skills mattered (using foundation models over building narrow ones), and what risks mattered. That is why the successor was a new plan, the National AI Action Plan 2026–2030, rather than a revision of the old one.
The timeline rests on the public record: the AI-Rmap’s 2021 publication (archived here), ChatGPT’s launch in November 2022, the 84% GenAI-usage figure cited by the Digital Minister at the September 2024 AIGE launch, and reporting that voluntary AI guidelines are under pressure from generative AI with legislation expected to follow.
The interpretation is independent analysis by airmap.my. Full sources: airmap.my/sources.