This article first appeared in Digital Edge, The Edge Malaysia Weekly on June 8, 2026 - June 14, 2026

Malaysia attracted over US$18 billion in artificial intelligence-related investment between 2022 and 2024. By the end of 2024, some 71% of Malaysian banking institutions had deployed at least one AI application, and the government’s AI Governance Bill is nearly complete. Every signal points to a country that has decided AI is central to its future.

And yet the question that will determine whether all of that investment actually delivers, the one question I rarely hear asked in any boardroom, is this: How connected, how current and how trustworthy is the data we are asking this AI to work with?

The answer to that question matters more than the technology itself, who provides it or how much is spent on it. Recent wars have shown the consequences of getting it wrong.

On the morning of Feb 28, the first day of the US-Israel military operations against Iran, a Tomahawk missile struck the Shajareh Tayyebeh girls’ elementary school in Minab during a morning session. At least 156 children and teachers were killed and 95 injured. The US military had used the Maven Smart System, an AI targeting platform capable of producing 1,000 target packages in one hour, with US forces striking 6,000 targets in Iran during the first two weeks of the operation.

Internal US investigations reported by The New York Times suggest the strike resulted from the school being incorrectly identified as a military base due to outdated data. The school had previously been adjacent to an IRGC naval facility, but the two had been separated years before the strike.

A preliminary inquiry concluded that the school was incorrectly identified as a military base during the targeting process. The internal report described it as a human error, not an AI malfunction.

The failure was not in the algorithm, but in relying on outdated data. By the time reality shifted, course correction was impossible.

I do not want to overstate the parallel. The consequences of that strike are in a category of their own. But the same pattern exists in financial institutions, hospitals and government agencies across this region: systems operating on data that no longer reflects the environment they are reading. The consequences are less catastrophic, but no less costly over time.

When I talk to boards of directors and leadership teams about their AI investments, the conversation almost always focuses on outputs where the fraud was flagged, the credit risk scored and the compliance report generated.

These are visible and measurable, so they become the yardstick for success.

What rarely gets asked is what those systems are unable to see. Think of it like diet and exercise. No amount of exercise can fully offset a poor diet. Your body can only work with what you feed it, and the quality of the output reflects the quality of the input. You feel fine until you do not, and by then, the damage has often been building for years.

Enterprise AI works exactly the same way. The model processes whatever data you feed it and produces outputs with whatever confidence the inputs allow. And in most organisations, nobody has a clear picture of how connected, how current or how clean that data actually is.

A fraud detection model fed by four disconnected systems will miss patterns that only emerge when the data is viewed together. A credit scoring model trained on past repayment behaviour will produce confident outputs in an interest rate environment it was never designed for. A clinical AI working from incomplete patient records is not making better decisions faster. It is making faster decisions with less information than the clinician it is meant to support.

None of these failures announce themselves. The system keeps producing authoritative-looking outputs, and people gradually stop questioning them.

Coming back to the US-Iran war, the US Government Accountability Office warned in 2021 and 2022 that the Pentagon lacked the complete, accurate and standardised data needed for advanced analytics. Reviews repeatedly found that the technology worked as intended, but the data underpinning it did not. The same condition exists in many enterprise AI deployments across Malaysia today.

Adoption is not the same thing as readiness, and in a regulated environment, that distinction carries real cost.

Malaysia lost 5.04% of its gross domestic product to money laundering in 2024 despite spending US$1.95 billion on compliance. The signals that would have caught more of that financial crime existed somewhere in the system. The architecture to connect them and surface them in time largely did not.

Bank Negara assistant governor Adnan Zaylani said it plainly at the Malaysian Banking Conference 2024. “AI models can exacerbate biases and discrimination, especially in cases where the underlying input data is flawed or of unknown quality. Strong data governance and management systems are, therefore, of paramount importance when embarking on AI projects.”

The regulator’s message is straightforward. The model is not the risk. The data environment is. Financial institutions that deploy AI without first asking whether the data is connected, current and properly governed are not more protected than before. They are moving faster towards the same blind spots.

The real lesson from Minab is not about AI, but about what AI inherits.

The Maven Smart System inherited stale data that humans never updated, then moved at a speed that made the consequences impossible to catch before they arrived. The AI did not fail. It succeeded at exactly the wrong thing.

That pattern is more common in enterprise settings than many leadership teams would like to admit. The model gets deployed, adoption metrics get reported and a data pipeline that nobody owns continues shaping every output until something external exposes the problem.

Malaysia’s AI Governance Bill and Bank Negara’s responsible AI framework set the right standard. But no framework validates whether the data your AI ran on this morning still reflects the market your institution is actually operating in today. That work belongs inside the organisation, with someone who has both the authority and the mandate to act on what they find.

Kew Yoke Ling is executive director at KewMann, an artificial intelligence and big data analytics company that leverages behavioural science to predict and influence human behaviour

Save by subscribing to us for your print and/or digital copy.

P/S: The Edge is also available on Apple's App Store and Android's Google Play.