David Bombal

#550: Firewall Demo of Red Team vs Blue Team: Hacking Finance Apps with AI Chatbots

Feb 22, 2026
Guest
Ant Ducker
Ant Ducker, a network security engineer who runs red team vs blue team firewall demos, walks through attacking a finance app with an LLM chatbot and defending it. Short, punchy demos show prompt injection attempts, lateral movement in Kubernetes, Zero Day detection, AI-driven guardrails, micro-segmentation, and encrypted-traffic inspection with neural intent detection.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ANECDOTE

Finance App Grew Into A Chatbot-Connected Hybrid System

  • The finance app evolved from a multi-tier on-prem app to a hybrid with Kubernetes, a proxy, and an LLM chatbot.
  • Ant's red team was engaged after the chatbot appeared unexpectedly, triggering the full attack-and-defend demo.
INSIGHT

AI Red Teaming Finds LLM Weaknesses

  • AI Defense runs algorithmic red teaming: it uses AI to automatically launch thousands of prompt-injection tests against deployed LLMs to find weaknesses.
  • In the demo a GPT-3.5-turbo model blocked ~69–70% of attacks but AI Defense surfaced remaining attack surface for remediation.
ADVICE

Convert Red Team Findings Into Chatbot Guardrails

  • Turn red-team discoveries into runtime guardrails on the chatbot to prevent prompt injection, code leakage, and PII exfiltration.
  • Ant demonstrated generating policies from tests and then switching from monitor to block so the same prompt no longer worked.
Get the Snipd Podcast app to discover more snips from this episode
Get the app