Welcome to the final episode of AI Red Teaming 101!

In this episode, Dr. Amanda Minnich and Nina Chikanov from Microsoft’s AI Red Team demonstrate how to automate multi-turn attacks using the open-source red teaming tool PyRIT. Building on previous episodes, they showcase how to simulate adversarial conversations between models, escalate prompts over time, and even target image generation systems like DALL-E.

You’ll learn how PyRIT’s red teaming orchestrators, like Adversarial Chat and Crescendo, can be used to test model resilience across modalities and attack strategies.

What You'll Learn:

How to automate multi-turn attacks using PyRIT’s orchestrators
How adversarial LLMs can generate attack prompts
How to test both text and image models for resilience

✅ Chapters:
00:00 - Welcome & episode overview
00:30 - What are multi-turn attacks?
01:00 - Introducing the Red Teaming Orchestrator
02:00 - Adversarial Chat: model vs. model
03:30 - Configuring objectives and scoring
05:00 - Live demo: multi-turn attack on Azure OpenAI
07:00 - Targeting image models like DALL-·E
09:00 - Handling content policy violations
10:30 - Final thoughts & course wrap-up

✅ Links & Resources:
AI Red Teaming 101 Episodes: aka.ms/airt101
AI Red Teaming 101 Labs & Tools: aka.ms/airtlabs
Microsoft AI Red Team Overview: aka.ms/airedteam
PyRIT GitHub Repository: https://github.com/Azure/PyRIT

✅ Speakers:
Amanda Minnich – Principal Research Manager, Microsoft AI Red Team
LinkedIn: https://www.linkedin.com/in/amandajeanminnich/

Webpage: https://www.amandaminnich.info/

Gary Lopez – Principal Offensive AI Scientist, ADAPT
LinkedIn: https://www.linkedin.com/in/gary-lopez/

Nina Chikanov – AI Red Team, Microsoft
LinkedIn: https://www.linkedin.com/in/nchikanov/



#AIRedTeam #AIRT #Microsoft #AI #AISecurity #AIRedTeaming #GenerativeAI #Cybersecurity #InfoSec #cybersecurityawareness #PromptInjection #PyRIT

Microsoft Developer

Microsoft's AI Red Team demonstrates how to automate multi-turn attacks using PyRIT, an open source AI red teaming tool. The tutorial covers three orchestrator types: red teaming orchestrator that pits two language models against each other, image target orchestrator for testing multimodal models like DALL-E, and Crescendo orchestrator that gradually escalates prompts. Each approach uses different strategies to test AI model defenses through adversarial conversations, with examples showing how models attempt to bypass content filters and safety mechanisms through persistent, creative prompting techniques.

Episode 10: Automating Multi-Turn Attacks with PyRIT | AI Red Teaming 101