Fighting New Era Bandits Who Are Plundering Your AI - ML Application

Explore techniques used by adversaries to steal or extract valuable AI/ML models without breaching system security in this 54-minute webcast from RSA Conference. Dive into real-life practical examples and implications of these threats for enterprises, presented by Manojkumar Parmar, Program Director at Robert Bosch Engineering and Business Solutions Private Limited. Learn about query-response mechanism attacks and other clever methods employed by new-era bandits. Discover active and passive detection measures and defenses to protect AI/ML models. Gain insights into security threats across the AIOT cycle, model extraction attacks, and their taxonomy. Examine industry-level threats to AI investment, including the CVE-2019-20634 ProofPoint Evasion example. Understand the technical aspects of attack vectors and results from real environments, including stealing ML models from Web APIs. Explore patented protection mechanisms against model extraction attacks, including a 14-family defense mechanism and the AlShield Tool. Conclude with prioritization strategies using simplified scenario planning to understand future critical assumptions in AI/ML security.

Intro
Session Outline Agenda
Security Threats across AIOT Cycle
Internal Experiment: Stealing/Extracting Al Model (1/2)
Data is no more oil: Model Extraction Attacks Industry Level Threat to Al Investment
Model Extraction Attacks - Taxonomy Exponential Growth in Attack Surface and types
CVE-2019-20634: ProofPoint Evasion Real life example
Technical Deep Dive : Attack Vectors & Results
Model Extraction Attacks - Results from Real Enviornment Stealing ML models from Web APIs
Internal Experiment: Example Kill chain Leveraging MEA Evading Person Detection - Intelligent Security Camera System
Patented Protection Mechanism against Model Extraction Att 14 family of Defense Mechanisms - Detection of attack
Patented Protection Mechanism Performance Active Defense: Detection + Action (Class Swap)
Patented Protection Mechanism against MEA AlShield Tool Demo
Prioritization with Simplified Scenario Planning Understand future with critical assumptions Critical Assumption