AI Jailbreaks and Red Teaming

15min Snip

00:00

Play full episode

Summary

Transcript

Episode notes

The chapter explores powerful jailbreaks in AI models such as Crescendo and Master key, emphasizing the challenges in securing AI models against harmful or offensive outputs. It discusses the use of tools like Pirate to automate risk identification in Generative AI models and the development of a tool using three AI models for red team operations. The conversation also covers the multidisciplinary nature of Infosec in tackling AI security challenges and delves into the speaker's past in writing cyber security-themed books.

Mark Russinovich, Eric Boyd & Neha Batra join us to discuss the state of AI for Microsoft and OpenAI at Microsoft Build 2024. It’s safe to say that Microsoft is all-in on AI.

Leave us a comment

Changelog++ members save 14 minutes on this episode because they made the ads disappear. Join today!

Sponsors:

Cronitor – Cronitor helps you understand your cron jobs. Capture the status, metrics, and output from every cron job and background process. Name and organize each job, and ensure the right people are alerted when something goes wrong.
1Password – Build securely with 1Password - 1Password simplifies how you securely use, manage, and integrate developer credentials. Manage SSH keys and sign Git commits. Access secrets stored in 1Password. Automate administrative tasks. Integrate with third-party tools. Also, check out our INFRASTRUCTURE.md file for more details on how we do secrets with 1Password.
Neon – The fully managed serverless Postgres with a generous free tier. Neon separates storage and compute to offer autoscaling, branching, and bottomless storage.
Fly.io – The home of Changelog.com — Deploy your apps and databases close to your users. In minutes you can run your Ruby, Go, Node, Deno, Python, or Elixir app (and databases!) all over the world. No ops required. Learn more at fly.io/changelog and check out the speedrun in their docs.