Generative AI is transforming our world, but how secure, safe, and aligned is it, really? This interactive, lab-style workshop dives deep into the critical practice of AI red teaming. Forget passive learning – here, you'll roll up your sleeves and actively probe generative AI models to uncover their vulnerabilities, biases, and potential for misuse.

Drawing from real-world red teaming initiatives (like those at DEFCON and NIST) and established techniques, participants will:

• Engage in Simulated Red Teaming Exercises: Get hands-on experience testing AI models against various challenges.
• Experiment with Jailbreak & Prompt Injection Techniques: Learn and apply methods like social engineering, character adoption, encoding attacks, and typographic tricks to try and bypass AI safeguards.
• Analyze Model Responses: Collaboratively identify different types of vulnerabilities, from generating harmful content and misinformation to revealing unintended biases or security flaws.

This workshop moves beyond theory to practical application. You'll gain first-hand insight into why red teaming is crucial for AI safety, application security, and platform integrity. We'll explore how it has evolved from military strategy and cybersecurity to become an indispensable tool for evaluating frontier AI models.

Who is this for?
Technologists, developers, researchers, policymakers, students, ethicists, and any curious digital citizen interested in understanding the practical challenges of making AI safer and more trustworthy. No prior red teaming experience is required, just an inquisitive mind!

What you'll leave with:

• Practical experience in basic AI red teaming techniques.
• A deeper understanding of AI vulnerabilities and how to identify them.
  Insights into designing red teaming exercises.
• A framework for thinking critically about AI safety, ethical implications, and the urgent need for robust evaluation and safeguards in the age of generative AI.

Come with your laptop connected to the internet, if you can!