Trusted Computing

Goals

• Understand the AI systems that need to be secured
• Analyze the ways AI helps and hurts security
• Connect the changing AI landscape to classic security principles

The AI Landscape

• What are the distinguishing characteristics + strengths/weaknesses of different forms of AI?
  • Learning
  • Reasoning
  • Search
  • Physical
  • Probabilistic
  • Optimization
  • Hybrid

AI Anatomy

• What are the stages in the training stack?

• What is a context window, and why does it matter?
  • What are the security implications?
• At a high-level, how does an agent work?

Securing AI

• What are the three axes in an AI threat model?
  • What are the options along each one?

• What are the components of the classical ML attack taxonomy?

• What are the vulnerabilities/attacks within the training stack?

• What is an adversarial example, and how does it affect security?

• How do agents change the security risks for AI?
  • What are some potential defenses?

Offense Using AI

• How has offensive use of AI evolved?

• What are the implications of Mythos-level models for security?

Defense Using AI

• How can AI help in…
  • Software security?
  • SOC operations?
• Why is AI slop a security problem?