AI GUIDE FOR BOARDS IN SINGAPORE
Section I: The Board Mandate
2.5 Potential Challenges
2.5.2 Technology and operational threats
Every new innovation inevitably has a double edge. Similarly, boards should keep an eye on the potential new challenges arising from AI.
AI failures can quickly become operational, legal, and reputational issues. Opaque decision-making, supply chain vulnerabilities, data leakage, and dependence on a small number of AI providers can undermine resilience and accountability. • Opaque architecture and explainability gaps: Frontier models and deep neural networks inherently lack mechanistic transparency, making it nearly impossible to trace exactly how an AI arrived at a speci c decision. e “black box” dilemma creates severe regulatory and operational anxiety for corporate boards; you cannot govern what you cannot see. Without a clear audit trail, boards face immense liability regarding unmapped algorithmic bias, hallucinated data and compliance violations under emerging AI safety legislation. When an autonomous agent makes a catastrophic nancial or legal error, the lack of explainability leaves leadership completely unable to defend the decision-making process to regulators, shareholders, or courts. • Hyper-sophisticated engineering and supply chain vulnerabilities: Frontier models lower the barrier to entry for highly sophisticated cyber warfare. e risk shi s heavily to the third-party supply chain and open-source so ware libraries integrated into the enterprise architecture. An attack on a minor vendor can instantly compromise the parent rm.
2.5.1 Frontier threats
T he threat landscape has changed dramatically with the introduction of frontier AI models, the most advanced general- purpose AI models at any given time. Frontier AI models represent a new class of AI-powered vulnerability discovery systems. Testing by the UK AI Security Institute found that these models can operate without human guidance to execute highly complex, multistep cyber attacks. Frontier AI models can successfully map thousands of long- standing, unknown “zero-day” aws across major operating systems and browsers at machine speed. While export controls and safeguards are actively debated, the genie is out of the bottle. Attackers can now have access to “asymmetric oence”, where a single hacker or small syndicate can breach major corporate or government networks. Traditional corporate patch-management cycles that take weeks are completely obsolete when an AI-driven exploit can weaponise a aw in hours. Boards must shi from a compliance- focused security mindset to an active, machine-speed resilience strategy.
28
Made with FlippingBook Digital Proposal Creator