Report 08 Aug 2024

The Danger of Adversarial AI as a Cyber Threat - InSight Analysis - Worldwide

PAC firmly believes and advises that the challenges and opportunities presented by adversarial AI necessitate a strategic and integrated approach to enhance the security and effectiveness of machine and deep learning systems.

 

Contents:

Management Summary

  • PAC’s opinion
  • PAC’s recommendations

What is adversarial AI?

How does adversarial AI impact generative, responsible, and explainable AI?

Who is using adversarial AI, how, and to do what?

What are the different types of adversarial attack vectors?

  • Evasion Attacks
  • Poisoning Attacks
  • Model Inversion Attacks
  • Membership Inference Attacks
  • Trojan Attacks
  • Model Extraction Attacks
  • Gradient-Based Attacks
  • Black-Box and White-Box Attacks

What is adversarial defence, and why is it essential?

  • Types of Adversarial AI Defence
  • Implementation in an Operational Context
  • Comprehensive Defence Strategy Implementation
  • Benefits of Adversarial AI Defences

What approach to governance supports adversarial AI?

What are the positive aspects of adversarial AI?