AI GUIDE FOR BOARDS IN SINGAPORE
Section III: Governance and Oversight
Illustrative Risk Scenarios
Scenario #1: Bias in AI-Assisted Workforce Decisions Scenario: An organisation pilots an AI tool to support workforce-related decisions, such as candidate screening, promotion recommendations or performance analysis. e system is trained or calibrated using historical organisational data that reects past workforce patterns and decision-making practices. Potential failure mode: Without appropriate controls, the AI system may reproduce or amplify historical bias, causing certain groups of employees or candidates to be unfairly disadvantaged. If the organisation treats the system’s outputs as objective without testing for bias, explainability and contestability, awed recommendations may be embedded into high-impact people decisions. Board considerations: AI systems used in workforce or other high-impact decision processes should be subject to rigorous bias testing, data-quality review, human oversight and clear escalation mechanisms. Boards should ask management how datasets are vetted, how outcomes are monitored across aected groups, and how individuals can challenge or seek review of AI-assisted decisions.
Scenario #2: AI- Accelerated Cyber reats Scenario: Advanced AI and AI-assisted so ware tools can help enterprise security teams identify vulnerabilities, analyse code, accelerate patching and improve cyber defence. e same broad class of capabilities may also be misused by malicious actors to scale reconnaissance, automate parts of vulnerability discovery, or accelerate exploit development. Potential failure mode: An organisation continues to rely on traditional cyber risk assumptions, such as annual penetration testing, static patching cycles or manual threat monitoring, even as attackers increasingly use AI-enabled tools to increase the speed and scale of cyber activity. is may leave the organisation exposed to faster-moving threats, delayed remediation and inadequate incident response readiness. Board considerations: Boards should ensure that cyber resilience strategies are updated to reect AI- accelerated threat activity. Management should review
Scenario #3: Accountability for Customer-Facing AI Channels Scenario: An organisation deploys an AI-enabled customer service tool to answer questions about products, services, eligibility criteria or customer entitlements. A customer relies on the tool’s response, but the information provided is inaccurate or inconsistent with the organisation’s o cial policies. Potential failure mode: e organisation may assume that disclaimers, website terms or alternative information elsewhere are su cient to limit its responsibility. However, where an AI system is presented as an o cial customer-facing channel, inaccurate or misleading outputs may create legal, regulatory, reputational and customer trust risks. Board considerations: Customer-facing AI systems should be governed as o cial enterprise communication channels. Boards should ensure management de nes clear boundaries for what such systems can represent, conducts pre- and post- deployment testing, monitors model performance, and provides clear escalation routes to human support when uncertainty or anomalies arise.
whether vulnerability management, patching
timelines, access controls, security monitoring, vendor risk management, red- teaming and incident response plans remain t for purpose in an AI-enabled threat environment.
61
Made with FlippingBook Digital Proposal Creator