I Hacked My Own AI Agent (Here's How You Can Too) | Ep 6

Most teams ship AI agents without ever testing if they can be broken. In this episode, I red team my own production AI agent live — prompt injection, jailbreaks, data exfiltration attempts, and tool-calling exploits — so you can learn to do the same before an attacker does. 📂 Materials & Code https://github.com/data-guru0/PYRIT-U... Follow me on LinkedIn here:   / sudhanshu-gusain976   Checkout the other episodes here:    • PRODUCTION AI SYSTEMS ENGINEERING SERIES   Timestamps: 00:00 What is AI Red Teaming ? 17:06 PYRIT Red Teaming 38:50 Targets, Converters and Scorers 01:07:20 Direct Prompt Injection 01:19:34 Multi-Turn Attacks 01:41:47 Encoding Attacks 01:50:46 Red Teaming Orchestrator 01:59:10 XPIA Attacks on AI Agents Data 02:22:50 Skeleton Key and Persuasion Attacks 02:34:02 Scoring Pipelines 02:40:10 Multimodal Attacks 02:52:44 Bulk Attacks and Temperature Fuzzing 03:09:31 PYRIT Red Teaming Dashboard App