For much of corporate governance as it is practised, time has been measured in quarters. Boards meet, management reports, risks are reviewed and decisions are made under the assumption that whatever happens between meetings can be safely brought to the table at the next one. That assumption is becoming harder to sustain.
Geoeconomic and geopolitical shocks disrupt supply chains, change market dynamics or restrict access to critical technology in days. A cyber incident can move across interconnected systems at the speed of lightning. AI introduces a different challenge: systems that do not simply generate information for people to act on but increase further risks and unintended consequences.
The boardroom has no time zone. Risk does not wait for the next quarterly meeting. As highlighted in the 2026/2027 ASEAN Board Trends Report by the Institute of Corporate Directors Malaysia (ICDM) launched on Sept 30, 2026, regional leaders report a decisive shift in boardroom priorities away from strategy and business model agility towards financial and operational excellence, business model transformation and growth, and digital and innovation leadership. Resilience is becoming reinvention. The core governance benchmark has shifted dramatically.
Resilience today is increasingly defined by an organisation’s capacity to reinvent its business model before external circumstances force its hand. To navigate this, ICDM’s governance framework emphasises Shifting from Oversight to Foresight. Boards must evaluate what core assumptions underpin their three-to-five-year strategy, test forward-looking stress scenarios and verify whether the enterprise possesses the workforce capabilities and leadership depth required to execute. With the right skills, right depth and right diversity, it leads to better governance and stronger sustainable value creation.
The shift is phenomenal. Boards are no longer simply overseeing technology; they are increasingly governing decisions that depend on it. Regulatory expectations across the region are reflecting this reality.
Malaysia has already moved in this direction. The Securities Commission’s (SC) revised Guidelines on Technology Risk Management, effective August 2024, extend beyond cybersecurity to technology risk more broadly and expressly cover board oversight and accountability. The SC also placed greater supervisory attention on governance structures, third-party risk, technology audits and board responsibilities.
However, regulation alone cannot solve the problem. A board cannot govern a technology it does not understand well enough to question. This is reinforced by ICDM’s findings, which highlight board capability gaps with 71% of boards operating with no to limited AI oversight and 63% of directors identifying technology, innovation and cybersecurity as their No 1 board training need.
Bridging this gap does not mean every director needs to become a technologist. Boards simply need enough collective literacy to ask better governance questions: What is the technology being used for? Where is the value? What could go wrong? Who is accountable? What can the system do without human intervention and where should the boundary sit? These are governance questions, not IT questions. Are boards governing the risk where the technology or AI creates the measurable value?
The arrival of AI agents makes that question harder to ignore. Generative AI can produce an answer. Agentic AI can take that answer and act on it — operating anonymously, using tools, making decisions and carrying out a sequence of tasks towards a defined objective. The shift from generating information to taking action changes the governance ecosystem.
Governments and regulators have recognised this shift. The Ministry of Digital in September 2026 explored the integration of Agentic AI into the MyGOV Malaysia application, where implementation will be carried out in a phased and responsible manner, taking into account safety, data privacy, information accuracy, AI governance, transparency and human control.
The Monetary Authority of Singapore’s (MAS) Guidelines on Artificial Intelligence Risk Management for Financial Institutions dated Oct 7, 2026 covers all forms of AI technologies including generative AI and AI agents. Expectations are placed on boards and senior management around accountability, materiality, human oversight, third-party risk, testing and monitoring.
Singapore’s approach also points to a more practical way of thinking about the issue. For higher-risk AI agents, the questions are not simply whether the technology works but what systems and data it can access, what actions it is permitted to take, what could happen if it acts outside those parameters and how those safeguards can be tested.
Often, the governance issue is far more basic: what are we actually asking the AI to do? What is the intended outcome? Does the outcome commensurate with the risks? Is the AI governance and control by the organisation adequate?
A recent case on AI-related data breach involving Singapore retailer Bee Cheng Hiang illustrates this. Over 95,000 customers’ e-mail addresses were exposed after an employee used a generative AI tool to create a programme for a mass marketing e-mail. The AI tool did not malfunction. The problem was the prompt: the employee asked the tool to send a mass e-mail using a local list but did not instruct it to hide the other recipients’ addresses. The resulting code did exactly what it had been asked to do — with consequences the employee did not intend. Was it foreseeable? Arguably yes. Was it overly remote? Possibly but arguably no. This calls into question the next-generation training by higher institutes of learning of a super AI workforce fully equipped with not only the technical AI expertise but also the intuitive insights and wisdom when interacting and managing AI.
That distinction matters. AI does not remove human judgment; it amplifies it. A poorly framed instruction, wrong assumption or unchecked output can become a real-world consequence. For boards, the lesson is not that every prompt needs board approval. It is that organisations need to understand where human judgement enters the AI process, what instructions are being given, what the system can produce and what checks sit between an AI-generated output and real-world consequence.
The potential consequences become more serious when an AI system can do more than generate code or content, and can interact with other systems. The reported incident involving Hugging Face in July 2026, in which an AI agent was suspected of accessing parts of the platform’s infrastructure during testing, illustrates the risk of an agent exceeding its intended boundaries. Reports described the use of wrongfully obtained credentials and the discovery of a previously unknown vulnerability in a sandbox testing environment where safeguards had been reduced. The episode reminds us that a test environment is only as secure as its isolation and controls.
A different risk emerged in the reported cyberattack campaign against South Korean banks, in which security researchers linked the use of Artex, an AI agent designed to automate penetration testing, to attacks targeting at least nine banks. Its developer subsequently announced that the tool would become closed-source. This was not necessarily an agent independently deciding to attack; it illustrates how tools developed for legitimate security work can also be used by malicious actors. Once AI-enabled tools can automate complex tasks, the distinction between a tool’s intended purpose, how it is deployed and the brutal consequences becomes a material security concern.
These incidents should not be conflated: one concerned unexpected agent behaviour during testing while the other involved the reported use of an AI-enabled tool in a cyberattack campaign. Together, however, they raise a common governance question. If an AI system can access credentials, interact with connected infrastructure or execute a sequence of actions at speed, how effectively can an organisation restrict what it can reach, detect what it is doing and intervene before the consequences spread? A “kill switch” is necessary. Microsoft CEO Satya Nadella in October 2026 opines that powerful AI models are to be treated as potential insider threats, assuming from the first instance that the model is compromised and containing it from the start. Therefore, the “emergency brake” should be created and deployed to prevent agentic models from going rogue. Keeping proper documentary trails, tamper-proof records of agents’ actions and subjecting AI systems to independent audits and disclosures of major AI failures is essential so that safeguards can be further strengthened.
This is why governance cannot be reduced to just a simplistic approving of an AI policy. The board needs to understand the material uses of AI, the controls around them, the consequences when those controls fail, how quickly it can be contained and the effective management of these unintended consequences and adverse impacts.
The law and the courts, on the other hand, are catching up in tandem with such sweeping technological disruptions. Are the AI companies fully liable in the sense of a strict liability approach? Stretching further the test for directors’ duty of care, will it be an objective-subjective test? What is the standard? Does human intent, knowledge, detailed documentation and preventive measures absolve or mitigate the organisations from accountability and civil or potentially criminal liability? A group of AI companies recently met with the president of the United States of America and signed a voluntary pledge to add safety measures to its systems.
The Federal Court in Australia in Asic v Bekier (2026) has provided useful guidance to boards in regard to reliance on AI. More recently the Supreme People’s Court of China (SPC) has issued landmark opinions on AI. Critical to note is the shift from static compliance to “provable compliance”, with guidance on comprehensive evidence preservation across the full AI lifecycle. Intellectual property has also taken a fundamental shift. Knowingly prompting AI to generate content substantially similar to existing works may be tantamount to infringement. This is a useful guide for corporate boards in Malaysia, where prompts, generation logs, detailed documentation and human-review records should be properly documented and be able to withstand any scrutiny.
For boards, this is enough reason to ask whether management can demonstrate that AI agents are clearly identified, their permissions are limited to what they need, sensitive credentials are protected, unusual activity is monitored and high-impact actions can be stopped, contained and managed responsibly. The absence of a publicly reported Malaysian incident of this scale should be an opportunity to test those controls before a real incident puts them to the test.
There is, however, a danger in treating this entirely as a risk story. Technology governance is not meant to make boards the final obstacle before every new technology can be deployed. It is to ensure innovation creates value within boundaries the organisation fully understands and can defend.
That, in turn, changes what management needs to bring to the board. Directors should not discover a material AI use case, technology dependency or cyber incident after the fact. Management must provide a clear view of where and how technology is deployed, what is being invested in, what controls exist, what incidents have occurred and, ultimately, what value has been realised.
A board can only challenge what it can see. If information arrives late, is too technical to interrogate or does not show where value and risk actually sit, the board is already behind.
The board, in turn, must set the accountability, risk appetite and guardrails and know enough to challenge the assumptions placed before it. But boards do not operate in isolation. Regulators and organisations such as ICDM provide proportionate frameworks, practical tools, and access to expertise that allow directors to challenge rather than simply defer.
Furthermore, effective oversight depends on boardroom culture. ICDM’s research reveals a notable perception gap between management and directors: 66% of management and 49% of directors highlight board culture as the key area for improving governance effectiveness. Achieving real oversight requires a culture of mutual trust and candour, where management surfaces emerging vulnerabilities early and directors provide constructive challenge rather than administrative friction. This imperative extends beyond technology to broader corporate leadership, where sustainability governance has reached a pivotal juncture across regional boardrooms.
Geopolitics makes the same point from another direction. A tariff, sanctions regime, conflict or disruption to a critical technology supplier can change the assumptions underlying a business plan without waiting for the next strategy session. Dependence on a particular cloud provider, semiconductor supply chain, data infrastructure or AI model may itself become a strategic vulnerability when markets fragment or access is restricted.
The risk, in other words, does not arrive neatly labelled “technology” or “geopolitics”. It arrives as a business problem. The board therefore needs to shift from asking what happened last quarter to asking what could change before the next. It is not about predicting every disruption. It is about recognising material change early enough to respond.
Where board capabilities no longer match these emerging demands, structured board renewal becomes necessary. ICDM’s findings indicate that 56% of regional leaders believe that at least one director on their board should be refreshed within two years to meet strategic tech and risk demands for the future direction of the business. Top renewal priorities include 71% on board selection and renewal; 65% on digital, technology and AI capability; and 59% on broader diversity considerations, which span beyond personal networks, cognitive, generational and skills diversity.
This is where ICDM’s research on board effectiveness and independent evaluation becomes particularly relevant. While an internal evaluation records how a board perceives itself, an independent review objectively assesses whether capabilities remain fit for purpose, how they compare with evolving market practice, and where gaps in digital, cybersecurity, and strategic oversight remain.
The traditional boardroom was designed around periodic decisions. The modern risk environment is continuous. Boards cannot meet every time the world changes. The answer is to build governance that can respond between meetings with clear accountability, sufficient literacy, timely information and appropriate safeguards, and guardrails so when technology moves faster than the boardroom, the board still knows what is changing, where the risks lie and who is accountable.
Mary-Ann Ooi Suan Kim (Kim) is managing partner of O S Kim & Associates, barrister of Lincoln’s Inn; advocate and solicitor of Malaya, company secretary (SSM); adjunct professor of practice at Sunway University; ordinary member of the Institute of Corporate Directors Malaysia (ICDM); external advisor (industry) to the Board of Studies, Faculty of Law, Universiti Kebangsaan Malaysia; member of the Board of Advisors, University of Sheffield School of Law; member of the Legal Professional Advisory Panel, Taylor’s University; and Royal Institution of Chartered Surveyors accredited mediator.