Practical AI Red Teaming - A Facial Recognition Case Study

Explore practical AI red teaming techniques in this conference talk focusing on facial recognition systems. Dive into the details of a real-world engagement testing both software and hardware solutions to identify critical vulnerabilities. Learn about the creation of an attack taxonomy and evaluation of recent approaches to compromising facial recognition algorithms. Discover insights from research conducted in authentic environments using various cameras and algorithms. Gain understanding of the prevalence of facial recognition technology, its vulnerabilities to adversarial attacks, and the cybersecurity implications. Examine the effectiveness of different attack methods, both digital and physical, against facial recognition systems. Consider defensive strategies and approaches to securing AI throughout its lifecycle. Benefit from the expertise of Alex Polyakov, a trusted AI and cybersecurity expert, as he shares findings from over 18 years of practical experience in the field.

Intro
Alex Polyakov
Adverse AI
Agenda
Why Secure AI
Confidentiality Integrity Availability
AI Applications
Who is affected
History of AI attacks
Top 10 AI attacks
Real applications
Real attacks
AI Red Teaming
Report
Air teaming
Problem
Attack Goal
Attack Form
Attack Actor
Attack Conditions
Attack Methods
Success Criteria
Results
Home Task
Digital Attack
Physical Facial Recognition
Goals
Existing research
Why test in the real environment
Device features
Approaches
Tricks
Example
Result
Defenses
The biggest problem
Highlevel approaches
Secure AI lifecycle
Next steps
Conclusion