AI Security

Taken two ways:

• Using AI to secure your product
• Using third party AI in your product securely

Securing AI

Red Team

• Prompt Injection
  • Repeating Word Prompt Injection
  • LLM Jailbreaking
  • Hypothetical Scenario Prompt Injection
  • Frog-Boil Attack
• ASCII Smuggling
• Slop Squatting
• Keras Deserialization Attack
• AI Hallucinations
• Copilot Cloud Data Labelling
• LLM Training Data Poisoning
  • MNIST Input Data Poisoning
• Model Extraction
• Evasion Attack
  • SecML
• Byzantine Attack
• Data Inference
• Model Inversion
• Membership Inferrence
• Glitch Token

Blue Team

• Canonical Form
• Polkadots Against Computer Vision
• Deep Neural Rejection
• SysVec
• System Prompt Canary
• AI Firewall
• Probe

AI For Security

• Ghidra MCP

Concepts

• AI Security Stages
• AI TRISM
• OWASP Top 10 for LLM

Talks

• The End of Click-Ops Security Agent-Driven Applications
• Asenion Security for AI