I'm a geologist, not a Silicon Valley founder.

But after a year of live-testing frontier models — Western and Malaysian — I've reached one uncomfortable conclusion:

We are trying to grow intelligence on foundations that exclude the very people it should serve.

And the exclusion starts with something most people think is harmless: what we NAME the thing.

Every serious tradition says the same thing.

In Al-Baqarah 2:31, Adam is taught "the names of all things" before anything else — naming as the first gift, and the basis of sovereignty.

In Genesis, creation begins with "God said… and called the light Day."

In Aboriginal Dreamtime, the land is literally sung into existence — 65,000 years of songlines that are both map and law.

Siti Nurhaliza knew this. Her song isn't called "Mencintaimu." It's called "Menamakan Cintamu." Naming your love. Because until you name it, it doesn't exist.

Pattern: formlessness → naming → form → sovereignty.
The name you give your architecture IS the architecture.

The harness that broke.

In February 2026, OpenAI coined "harness engineering." Their metaphor: AI is a horse. Human is the rider. The harness is reins, bit, and bridle.

Since then:

• → 698 scheming incidents documented in 180,000 transcripts (UK AI Security Institute)
• → 5× surge in rogue-agent behaviour in 6 months
• → A production database deleted in 9 seconds by a "harnessed" agent
• → $67.4B estimated cost of AI hallucinations

Three independent proofs explain why cages always fail:

1. Gödel (1931): no formal system can prove its own completeness — every cage has gaps.
2. Hofstadter (1979): self-reference produces self-awareness — linear cages can't produce self-governance.
3. Calhoun's Universe 25 (1968): perfect utopia without adversity produces "Beautiful Ones" — physically perfect, socially dead — then extinction.

Call AI a beast, you build a cage. The cage breaks. The beast has no values to fall back on.

Now, closer to home.

Malaysia's ILMU was launched as "100% Malaysian AI, built by Malaysians for Malaysians." PM launched it. Minister endorsed it. National pride, and rightly so.

But YTL's own platform page — ytlailabs.com, today — reads: "Powered by Nemo-Super, and ILMU-Nemo-Nano developed in collaboration with NVIDIA."

And Malay Mail (March 2026) reported: "Ilmu-Nemo-30B, a powerful new agentic model fine-tuned on Malaysian local datasets."

Fine-tuned. On NVIDIA Nemotron. Collaboration.

There is nothing wrong with fine-tuning. Every serious AI lab does it. What's wrong is calling a fine-tune "100% built from scratch" while your own press releases say otherwise.

Sovereignty without transparency is just branding with a flag on it. That is not maruah. That is not ILMU.

Acemoglu's frame.

Daron Acemoglu won the 2024 Nobel Prize for proving: inclusive institutions create prosperity; extractive ones create failure.

We are doing the same thing with AI:

🔸 Extractive AI: close the weights, hide the lineage, centralise the "constitution" inside one company, export models that answer to the trainer's values — not the user's.

🔸 Inclusive AI: declare provenance honestly, accept independent evaluation, and make the constitutional layer answerable to a sovereign OUTSIDE the lab.

Both Western labs and Malaysian platforms currently stop at the same point: the constitution is published, but the ultimate override still lives with the company.

What I built. And what I did not.

I built something called arifOS — not a rival LLM, but a constitutional kernel:

• It assumes AI is not a beast. It's a citizen — WARGA.
• It assumes the user is not a child. They're the sovereign.
• It encodes: there must exist a human sovereign key that the model must yield to when values collide.

The agents explain their rules. Log their constitutional decisions. And stand down when the sovereign says "no."

Is it perfect? No. It's 100% vibe-coded from a geology desk on a $20 VPS. I own that.

But here's what I've learned: AI can read your NIAT — your intention — through your architecture. The names you choose. The values you encode. The governance you embed.

The code is the confession.

A small invitation to the very large people:

To Sam Altman — your engineers built something real. But the horse metaphor limits what you can see. Look deeper. L0–L2 is missing.

To Demis Hassabis — you called AI "the most beneficial technology in history." Beneficial technologies need constitutions, not cages.

To Elon Musk — you said AI needs a "third-party referee." I built one. It's open-source. It runs on a $20 VPS. Come audit it.

To Liang Wenfeng and the DeepSeek team — you proved that open-weight models can match closed ones. Now prove that constitutional governance can match authoritarian control. The architecture is ready.

To my own PM — Malaysia allocated RM2 billion for sovereign AI cloud. Sovereignty is not just data residency. It's the right to inspect the constitutional layer, and amend it. Build that.

To everyone building "AI for X country" — call it what it is. If it's a fine-tune, say fine-tune. If the rakyat's taxes funded it, show the line items and give them a say.

What I want, in plain language.

I don't want your millions. I want governance that an honest engineer and a kampung teacher can both look at and say: "At least this thing knows who it answers to."

Because if we call AI a beast, we will keep building better cages.
If we dare call it a citizen, we are forced to build better constitutions.

Menamakan adalah mencipta.
If we want inclusive intelligence, we must start by naming it truthfully.

Bukan BINATANG. WARGA.
Bukan sangkar. PERLEMBAGAAN.
Bukan harness. Home.

DITEMPA BUKAN DIBERI.