CPTEnglishBeginner· Artificial Intelligence, Cybersecurity

AI Under Attack: Adversarial Inputs, Prompt Injection, and Model Security

Master the art of defending AI systems against adversarial attacks and manipulation

1h 30 lessons

What you'll learn

Identify the unique vulnerabilities that distinguish AI models from traditional, deterministic software.
Analyze how adversarial inputs and small perturbations can force unintended model behavior.
Explain the mechanics behind training data poisoning and strategies for maintaining data provenance.
Evaluate the risks associated with prompt injection and unauthorized tool execution in AI agents.
Apply defensive design patterns to secure the entire AI development and deployment lifecycle.
Detect hidden backdoors, latent triggers, and sleeper behaviors engineered into models.
Implement robust monitoring systems to identify abnormal model behavior and potential security incidents.

About this course

This course examines the security architecture of artificial intelligence, focusing on the specific risks that probabilistic models introduce to financial institutions in Hong Kong. As firms integrate machine learning into trading platforms, automated advisory services, and risk management tools, they face unique challenges that fall outside the scope of traditional software security. For SFC-licensed professionals, understanding these vulnerabilities is essential for fulfilling their obligations under the Management, Supervision and Internal Control Guidelines, as AI integration requires robust operational resilience and risk oversight. This content provides the technical grounding necessary to identify how adversarial inputs and prompt injection can bypass existing security controls.

The curriculum explores the transition from traditional deterministic cybersecurity to the management of nondeterministic AI systems. It addresses the expanded attack surface created by modern AI ecosystems, including data ingestion, training pipelines, and deployment environments. By examining the mechanics of model poisoning, latent backdoors, and inference-time manipulation, participants will gain insight into how these threats intersect with the regulatory expectations for systems integrity and data privacy. The course highlights the importance of data provenance and quality controls as a baseline for maintaining reliable AI output in environments handling sensitive financial records.

Finally, the course structure guides learners through the AI lifecycle, from initial development to the deployment of autonomous agents. It addresses the practical implementation of red teaming and adversarial testing to stress-test models against malicious probing. By focusing on architectural isolation and defense-in-depth strategies, the program equips professionals to evaluate AI risk in the context of broader organizational infrastructure. This training supports the ongoing need for institutional transparency and technical accountability when deploying automation, ensuring that professionals can navigate the ethical and governance requirements inherent in high-stakes financial technology adoption.

Who should attend

Compliance Officers in financial institutions
IT Risk Managers and Cybersecurity Specialists
SFC-licensed Responsible Officers
Fintech project managers and developers
Internal Auditors focused on technology controls
Digital asset strategy leads

Key topics covered

AI Security Fundamentals

Distinguishing AI model security from traditional software
Understanding the expanded AI attack surface
Principles of data integrity and model provenance
Managing systemic risks in complex AI supply chains

Adversarial Threats and Mitigation

Mechanics of training data poisoning and contamination
Detecting hidden backdoors and sleeper behaviors
Strategies for mitigating adversarial input attacks
Executing red teaming and adversarial testing

LLMs and Operational Governance

Identifying direct and indirect prompt injection risks
Managing security for autonomous AI agent tool execution
Resolving instruction hierarchy conflicts
Architecting systems for AI isolation and control

How this course satisfies CPT

This course covers Artificial Intelligence, Cybersecurity topics recognised by the SFC as eligible for CPT. Upon completion, you'll earn approximately 1 CPT hours and receive a certificate of completion for your training records. The course includes quizzes to verify your understanding, satisfying the SFC's requirement that online learning includes an independent assessment and maintains a reliable audit trail.

Learn more about CPT requirements

Curriculum · 30 lessons

1:02:15.2 of content

1.
Understanding the Critical Need for Securing Artificial Intelligence Systems
FREE
2:00.3
2.
Identifying Unique Cybersecurity Challenges Posed by Modern AI Models
FREE
1:56.7
3.
Differences Between Traditional Software Security and AI Model Security
FREE
1:59.5
4.
Foundational Principles for Securing AI Systems and Data Infrastructure
1:59.8
5.
Mapping the Expanded Attack Surface of Modern AI Ecosystems
2:07.1
6.
Quiz: Understanding the Critical Need for Securing Artificial Intelligence Systems – Mapping the Expanded Attack Surface of Modern AI Ecosystems
QUIZ
5 questions · requires lessons
7.
Securing the AI Lifecycle from Development to Deployment and Integration
2:03.9
8.
Mitigating Risks of Training Time Attacks and Data Contamination
1:58.1
9.
Implementing Data Provenance and Quality Controls to Prevent Poisoning
2:10.5
10.
Detecting Hidden Backdoors and Sleeper Behaviors in AI Models
2:06.0
11.
Managing Inference Time Risks and Adversarial Inputs During Deployment
2:06.8
12.
Quiz: Securing the AI Lifecycle from Development to Deployment and Integration – Managing Inference Time Risks and Adversarial Inputs During Deployment
QUIZ
5 questions · requires lessons
13.
Understanding the Mechanics and Dangers of Adversarial Machine Learning
1:56.5
14.
Addressing Model Vulnerability to Small Perturbations and Input Manipulations
2:07.6
15.
Securing AI Against Physical World Adversarial Attacks and Sensor Spoofing
2:07.4
16.
Achieving Robustness in AI Systems Against Intentional Adversarial Probing
2:08.3
17.
Implementing Effective Red Teaming and Adversarial Testing Strategies
2:08.9
18.
Quiz: Understanding the Mechanics and Dangers of Adversarial Machine Learning – Implementing Effective Red Teaming and Adversarial Testing Strategies
QUIZ
5 questions · requires lessons
19.
Addressing Prompt Injection Risks in Large Language Model Architectures
1:55.3
20.
Differentiating Between Direct and Indirect Prompt Injection Vulnerabilities
2:01.1
21.
Securing AI Assistants Connected to Sensitive Enterprise Data Sources
2:04.7
22.
Managing Security Risks Associated with Autonomous AI Agent Tool Use
2:02.1
23.
Resolving Instruction Hierarchy Conflicts to Maintain System Control
2:03.2
24.
Quiz: Addressing Prompt Injection Risks in Large Language Model Architectures – Resolving Instruction Hierarchy Conflicts to Maintain System Control
QUIZ
5 questions · requires lessons
25.
Preventing Model Extraction Attacks and Protecting Proprietary AI Assets
2:05.7
26.
Addressing Privacy Leakage and Data Memorization in Trained AI Models
1:57.0
27.
Securing Software Development Pipelines Against AI-Generated Code Vulnerabilities
2:09.9
28.
Managing Supply Chain Risks in Complex AI System Deployments
2:13.0
29.
Addressing Security Challenges in Multimodal and Embodied AI Systems
2:08.2
30.
Quiz: Preventing Model Extraction Attacks and Protecting Proprietary AI Assets – Addressing Security Challenges in Multimodal and Embodied AI Systems
QUIZ
5 questions · requires lessons
31.
Implementing Defense in Depth Strategies for Probabilistic AI Architectures
2:05.9
32.
Designing Secure Architectures to Isolate AI from Critical Infrastructure
2:00.8
33.
Establishing Standards and Accountability for Trustworthy AI Deployment
2:13.9
34.
Navigating Ethical and Governance Decisions in High-Stakes AI Adoption
2:09.1
35.
The Strategic Importance of Securing AI as Critical Infrastructure
2:08.0
36.
Quiz: Implementing Defense in Depth Strategies for Probabilistic AI Architectures – The Strategic Importance of Securing AI as Critical Infrastructure
QUIZ
5 questions · requires lessons

Requirements

A foundational understanding of cybersecurity principles and risk management.
Basic familiarity with how machine learning models and neural networks are trained.
Interest in AI safety, security architecture, or system reliability.
Comfort with high-level technical discussions regarding software development and data pipelines.

Certificate of Completion

Complete this course to earn a verifiable Certificate of Completion for your CPT training record. Here's how:

1Watch all the lessons — work through every lesson in the curriculum. Each completed lesson brings your progress closer to 100%.
2Pass all the quizzes — score 70% or higher in each quiz within the course. Quizzes are part of the curriculum and must be passed to finish.
3Download your certificate — once you reach 100% progress, download a personalised Certificate of Completion showing your name, the CPT hours earned, and your quiz score.

Sample certificate: