Trustworthy AI: Building Security Into AI Systems

Executive Snapshot

• Figure: Executive Snapshot: Trustworthy AI — Building Security Into AI Systems

Situation Overview

• The Emerging Security Risk to AI Systems

• AI in Cybersecurity

• Figure: AI in Cybersecurity

• AI Life Cycle and Model Security

• Figure: AI Definitions

• Figure: Security of AI Life Cycle

• Types of AI Model Threats

• Figure: Adversarial Threats

• AI Model Reconnaissance

• Poisoning Attack

• Evasion Attack

• Prompt Injection Attack

• Supply Chain Attack

• Privacy Attacks

• Model Replication

• Model Exfiltration

Advice for the Technology Buyer

• Defense Against AI Security Threats

• 1. Identify: Assess AI Security Risk and Posture

• AI Asset Mapping

• Use-Case–Based Risk Assessment

• 2. Protect: Implement Safeguarding Measures

• Security Awareness

• Model Safeguards

• Security by Design

• 3. Detect: Enable Timely Discovery of AI Security Events

• Security Monitoring

• 4. Respond: Prepare for AI Security Incidents

• Attack Response Plans

Learn More

• Related Research

• Synopsis