Mark Russinovich Talks Jailbreaks

Episode 21 of the Microsoft Threat Intelligence Podcast podcast, hosted by Microsoft, titled "Mark Russinovich Talks Jailbreaks" was published on June 19, 2024 and runs 31 minutes.

June 19, 2024 ·31m · Microsoft Threat Intelligence Podcast

0:00 / 0:00

Summary

On this week's episode of The Microsoft Threat Intelligence Podcast, Sherrod DeGrippo is joined by Mark Russinovich. Mark Russinovich, CTO and Technical Fellow of Microsoft Azure, joins the show to talk about his journey from developing on-prem tools like Sysinternals to working in the cloud with Azure. Sherrod and Mark discuss the evolution of cybersecurity, the role of AI in threat intelligence, and the challenge of jailbreaking AI models. Mark shares his experiences with testing AI models for vulnerabilities, including his discovery of the "Crescendo" and "Masterkey" methods to bypass safety protocols. They also touch on the issue of poisoned training data and its impact on AI reliability, while highlighting the importance of staying ahead in cybersecurity. In this episode you’ll learn: The shift from desktop computing to cloud-based systems and its implications Potential consequences of AI models having overridable safety instructions How AI training data can manipulate the outcomes generated by AI models Some questions we ask: Will AI owners be able to stop data poisoning, or will it become more common? Can you share challenges and vulnerabilities in maintaining the security of AI systems? What sparked your interest in AI jailbreaks, and what trends are you seeing? Resources: View Mark Russinovich on LinkedIn View Sherrod DeGrippo on LinkedIn AI jailbreaks: What they are and how they can be mitigated? https://www.microsoft.com/en-us/security/blog/2024/06/04/ai-jailbreaks-what-they-are-and-how-they-can-be-mitigated/ Inside AI Security with Mark Russinovich | BRK227 https://www.youtube.com/watch?v=f0MDjS9-dNw How Microsoft discovers and mitigates evolving attacks against AI guardrails. https://www.microsoft.com/en-us/security/blog/2024/04/11/how-microsoft-discovers-and-mitigates-evolving-attacks-against-ai-guardrails/ Google AI said to put glue on pizza. https://www.businessinsider.com/google-ai-glue-pizza-i-tried-it-2024-5 Related Microsoft Podcasts: Afternoon Cyber Tea with Ann Johnson The BlueHat Podcast Uncovering Hidden Risks Discover and follow other Microsoft podcasts at microsoft.com/podcasts Get the latest threat intelligence insights and guidance at Microsoft Security Insider The Microsoft Threat Intelligence Podcast is produced by Microsoft and distributed as part of N2K media network.