AI GUIDE FOR BOARDS IN SINGAPORE
Section V: Organisational Resilience and Capability
Secure AI Deployment
ZERO-TRUST ARCHITECTURE
ENDPOINT PROTECTION
Require strong authentication Apply rate limiting to prevent abuse Log and
Assume breach
Enforce least- privilege access
Encryption
Encrypt data in transit and at rest
monitor all API activity
Segment AI workloads from core infrastructure
SECURE AI DEPLOYMENT
Identity & Access Management
Administrative Controls
RED-TEAMING & ADVERSARIAL TESTING
TRAINING DATA PROTECTION
Classify sensitive data Restrict and control access Audit use and changes
Test AI systems against real- world attacks to identify vulnerabilities before deployment
Enterprise Privacy Protections
• Training data protection: Ensuring sensitive training data is classi ed, access- controlled and auditable. • Adversarial testing: “Red-teaming” and testing AI systems before deployment to identify vulnerabilities and failure points. AI systems should be governed with the same level of discipline applied to other enterprise- critical systems. Clear accountability, secure
Boards should request a consolidated view of AI-related cyber threats and how these risks map to existing enterprise security controls. Best practices include security controls across the AI lifecycle: data protection, model robustness, adversarial testing, incident response including real-time ltering for harmful content in generative AI outputs. where internal policies or regulatory requirements aect where data is processed or stored.
integration with internal systems and continuous monitoring are critical.
For organisations operating in Singapore and across Asia, data residency considerations may also form part of broader cyber resilience and trust discussions, particularly
92
Made with FlippingBook Digital Proposal Creator