Monday 16 Nov 2026
main news image

This article first appeared in Forum, The Edge Malaysia Weekly on November 16, 2026 - November 22, 2026

In July 2000, a company circulated a 64-page code of ethics. It spoke of respect, integrity, communication and excellence. Its foreword was signed by Kenneth Lay, its chairman and chief executive, and employees were instructed to sign a certificate agreeing to comply. Barely 1½ years later, Enron was bankrupt.

The principles were right. What was missing was the willingness to live by them when it became costly to do so.

Malaysia is now preparing a new framework for AI governance. On Oct 5, Digital Minister Gobind Singh Deo told the Dewan Rakyat that the Artificial Intelligence Governance Bill was expected to be tabled in early 2027. It is intended to take a risk-based approach across sectors, complementing existing laws. That is a welcome direction: enabling AI’s benefits while building the confidence needed to use it responsibly.

The following day, the inaugural Financial Lines Forum in Kuala Lumpur brought together those who make decisions in boardrooms and hospitals, those who advise and insure them, and those who examine what happens when decisions go wrong. AI accountability was a recurring concern. In the special address, Tan Sri Abdul Rashid Hussain, the founder of the RHB Group and chairman of Lembaga Tabung Haji, traced the evolution of corporate governance from rules to responsibility, beginning with that Enron code. I draw on his address here.

He observed that “by any measure, we have never had more laws, more codes, more disclosure and more scrutiny than we have today”. Yet every few years another inquiry, another investigation or another courtroom returns to the questions he posed: “How did the board not see it? Or worse: how did the board see it and say nothing?” Such failures, he said, rarely happen because there are no rules. “They happen in the spaces the rules do not reach.”

The July consultation on the AI Bill proposed national governance principles, central oversight and obligations proportionate to risk. These remain proposals and the months before tabling are a valuable opportunity to make them work in practice.

The consultation paper rightly states that responsibility for AI must remain with identifiable people, not with the AI system itself. The practical question is how institutions give that principle effect.

For companies deploying AI, that brings us to the boardroom.

AI already influences who obtains credit, which transactions are flagged and what customers and patients are told. Directors owe duties of reasonable care, skill and diligence under the Companies Act 2016. That does not make them personally answerable for every system error. It makes their oversight of material risks a proper subject of scrutiny.

Section 216 of the Companies Act offers an analogy. Directors who delegate a power remain responsible for its exercise as if it were their own, unless, at all times, they reasonably believed that the delegate would exercise the power in conformity with their duties under the Act and the company’s constitution, if any, and they reasonably and in good faith believed, after proper inquiry where needed, that the delegate was reliable and competent.

The section concerns delegation to persons, not systems. But the boardroom questions remain, namely, what power did you hand over, on what grounds and with what safeguards?

In early 2024, an employee at Arup’s Hong Kong office was deceived by a pre-recorded video conference that used deepfakes to impersonate senior management and authorised transfers of approximately 

US$25 million. Arup’s chief information officer later described the incident as technology-enhanced social engineering: its systems had not been compromised.

That same year, a Canadian tribunal held Air Canada responsible for misleading bereavement fare information supplied by its website chatbot. The airline could not treat the chatbot as independently responsible for what it told a customer.

The first concerns deception from outside the institution; the second, misinformation through its own system.

Section 17A of the MACC Act offers a warning from another field. When a commercial organisation commits the corruption offence specified there, its directors and officers are deemed to have committed it too, unless they prove that it was committed without their consent or connivance and that they exercised the due diligence required to prevent it, having regard to their functions and the circumstances. The organisation has its own defence of adequate procedures. The burden has shifted: “I did not know”, on its own, does not satisfy the statutory defence.

For AI, “the algorithm did it” should fare no better. What matters is how the system was selected, authorised and supervised, and what happened when concerns arose.

Directors need not become data scientists, Abdul Rashid said in his special address, and he put four questions to boards:

Where are we using AI?

Who is accountable for it?

 How would we know if it went wrong?

And could we explain its decisions to a regulator, a customer or a court?

Boards should ask them of management and insist on answers they can test.

Malaysia’s regulators are already addressing these questions. The Securities Commission Malaysia’s July consultation proposed clearer expectations for board oversight of technology and for listed companies’ disclosures on their use and governance of AI.

In his July 8 address to the banking industry, the governor of Bank Negara Malaysia said that AI “belongs firmly on the board agenda” and that “responsibility cannot be delegated to an algorithm”.

For boards and institutions using AI, six practical considerations follow.

First, name the people and give them the means to act

For high-risk deployments, an identifiable senior person with authority, resources and access to information makes accountability more effective. Naming that person should neither discharge the board nor make the individual the default bearer of blame.

Institutions also need their suppliers to disclose material limitations and changes and to provide records sufficient to investigate failures. Where adequate assurance cannot be obtained, use should be restricted.

Second, recognise responsible decision-making

Not every system that fails is a failure of governance and an honest, careful and informed decision that turns out badly is not, in itself, wrongdoing.

The Companies Act’s business judgement rule reflects that concern within defined limits. Its protection in relation to the duties of care, skill and diligence requires proper purpose and good faith, no material personal interest, an appropriately informed decision and a reasonable belief that it is in the best interest of the company.

Organisations that invest in adequate procedures, raise concerns early and put things right deserve recognition for it. If the AI Bill provides for a due-diligence defence, its conditions and scope will matter. An assessment filed away before launch should count for little if later warnings were ignored. A regulatory defence should not, by itself, extinguish an affected person’s civil claim.

Third, make human oversight mean something

In his keynote at the forum, Attorney General Tan Sri Mohd Dusuki Mokhtar asked how reliable an AI system’s output is, what the limits of its capabilities are, who is responsible for checking it and how a decision informed by AI should properly be documented. He stressed that effective human oversight remains essential.

Oversight requires competence, information, time and authority to intervene. A reviewer handed an unexplained recommendation and no realistic power to change it cannot provide meaningful supervision. The Hong Kong fraud underscores the need to verify apparent instructions independently.

For AI agents capable of acting, the questions become sharper: what may the system do, which actions require human approval and who can stop it? Material changes to its model, data, permissions or purpose require a fresh look.

Fourth, ensure that those governing remain capable of doing so

Fitness and propriety cannot be tested once and then assumed: a director must remain fit throughout the appointment. Nomination committees need to look beyond credentials to character, competence and a record of judgement. Technology literacy now belongs within that assessment.

Boards need enough understanding to challenge management and recognise when independent expertise is required. This applies equally to institutions holding a public mandate. Practical guidance, written for directors rather than engineers, would help.

Fifth, understand what insurance can and cannot do

As Abdul Rashid reminded the forum, a directors and officers (D&O) policy “can protect your balance sheet but it cannot protect your name”. No insurance product, he added, “has yet been designed that can buy back a reputation”.

In an analysis published on Oct 5 and drawn largely from US litigation, the broker Aon reported that D&O cover for AI risks remained broadly available while underwriters were examining governance more closely. Cover nevertheless depends on each policy’s wording, exclusions and endorsements.

The record of what a system was permitted to do, who approved it, how it was supervised and what happened when concerns arose may matter to an insurer, a regulator or a court. A board’s policy wording is a governance question and the time to examine it is before the claim.

Sixth, remember the person on the other side of the decision

Consider a small business owner refused financing after an AI assessment. The officer cannot explain which information mattered. The supplier invokes confidentiality. The applicant requests a review and the same system gives the same answer.

The refusal may be justified. The problem is that the applicant has no practical way to test it.

Those four questions therefore need a fifth, asked from outside: what can that person do?

The consultation paper’s first principle, on human dignity, already points the way: important decisions should remain open to human review. In practice, meaningful accountability includes a route to obtain useful reasons, correct inaccurate information and secure reconsideration by someone authorised to change the outcome. Thousands of lines of code would rarely provide the explanation the person needs.

The relationship between regulatory duties and civil remedies also matters, because a regulator’s power to fine does not necessarily compensate someone who has suffered loss.

Clarity on how the framework applies to AI used in public services would also be welcome, since some of the most consequential decisions affecting a citizen’s life are made by public bodies.

Legal safeguards, however, are only part of the answer. The rest is moral and it must be cultivated early. In parliament on Feb 24, the prime minister emphasised that ethical considerations would receive particular attention in drafting the Bill.

There is also a danger that lies in us rather than in the machine. As Dan Rust of Infopro Learning has written in CEO­WORLD Magazine, while we worry about AI becoming more like us, the real danger is that we are becoming more like it. An algorithm can calculate; it cannot judge what is right or relieve us of the responsibility for deciding.

Our traditions remind us that knowledge cannot be separated from character. Confucius taught that a young person should be filial, respectful, earnest and truthful, and devote remaining time and energy to study. The Malay-Islamic values of adab and amanah likewise remind us that knowledge carries obligations and is held in trust.

That formation begins in homes, schools and universities, long before anyone reaches a boardroom.

The question is already present whenever work is passed off as one’s own without acknowledgement or an answer is presented by someone who has not examined it. Education must cultivate the habit of asking, checking and taking responsibility.

Institutions must sustain that habit. They can adopt an AI policy while rewarding staff who overlook its weaknesses. They can appoint a reviewer while making disagreement professionally costly. They can produce minutes recording approval without showing that anyone challenged the assumptions.

Abdul Rashid closed his address with a reminder worth repeating: rules can require accountability, codes can demand transparency and structures can protect independence but no rule can manufacture integrity and no regulation can prescribe courage. He also left every board with a question: if its minutes were read aloud in court tomorrow, would they show that the right questions were asked or only that the directors were present?

The Bill is expected to be tabled early next year. Boards need not wait for it to begin answering that question. Legislation will set the framework; whether responsibility follows will be decided in the boardroom, one decision at a time.


JJ Chan is the founder of the Financial Lines Forum, managing partner of CBE (Chan Ban Eng & Co), barrister-at-law of Gray’s Inn, London, advocate and solicitor of the High Court of Malaya and an adjunct professor at the Faculty of Law, Universiti Malaya. He specialises in financial lines, medical negligence, corporate and directors’ liability, insurance/reinsurance and AI governance risk. He also writes regularly on constitutional and public interest matters.

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.

      Print
      Text Size
      Share