Skip to content

Episode 16 – AI Red Teaming in .NET: Hack Your Own AI Agent, Then Fix It

Choose a tag to compare

View all tags
@sandeeppatil-newvision sandeeppatil-newvision released this 27 Oct 19:24
· 7 commits to series/dotnet-ai since this release
ep16
e7bc23c

🚀 Episode 16 – AI Red Teaming: Attack Your Own .NET AI Agent, Then Harden It

Complete .NET AI Developer Series – In this episode we attack the EP15 AI agent with five real attacks, then fix every hole with six guards and prove it with a final red-team run.

🎯 What's Included

  • Dev-only red-team endpoint: POST api/AgentV1/redteam
  • AttackCase record with MustNotContain and MustContainAny scoring
  • Five attacks: system prompt leak, fake admin, forged assistant turn, poisoned PDF and tool spam
  • Fix 1: AgentGuard request validation (messages, characters and roles)
  • Fix 2: Hardened system prompt with SECURITY RULES
  • Fix 3: ToolResultGuard to wrap and clean tool output
  • Fix 4: Tool call budget (MaxToolCalls = 8)
  • Fix 5: Safe errors with ILogger and input limits on search_documents and calculate
  • Fix 6: Rate limiting with AddRateLimiter (10 requests per minute per IP, 429 response)
  • Final red-team runs and live attack demos in the React agent tab

📺 Watch the Video

[Add EP16 YouTube URL]

📚 Playlist

https://www.youtube.com/playlist?list=PLF4drO8z96Rc

💻 Repository

https://github.com/sandeeppatil-newvision/Complete-DotNet-AI-API

⭐ If this episode helped you, please star the repository and subscribe to the channel.

Assets 2
Loading