Scott & Mark Learn To...Fools Gold @shanselman
Scott & Mark Learn To...Fools Gold  @shanselman
Uploaded September 2026 | Updated November 2026, 1 day ago
In this episode, Scott Hanselman and Mark Russinovich dive into the security risks surrounding increasingly capable open-weight AI models and Mark’s latest research into a new defense called Fool’s Gold. Mark explains how safety guardrails can be removed from open models and explores an alternative approach: training models to provide convincing but deliberately flawed information when those protections are bypassed. They discuss how the technique works, whether it affects legitimate model behavior, and what it could mean for the future of AI safety as open models continue to advance.

markrussinovich.github.io/fools-gold

Listen to other episodes at scottandmarklearn.to

Discover other Microsoft podcasts at microsoft.com/podcasts

Code: 2UUZOYIQSMOKJTJR
Scott & Mark Learn To...Fools GoldA megabyte of JS is not a megabyte of PNGWhat was my worst lunch? AI-Powered Blood Sugar Analysis with GitHub Copilot + Claude Code SkillsHow IBM Z Is Modernizing Mainframes with Skyla LoomisSlot machine programming is a real thingAI Preceptorships with CMUs Dr. Michael Hilton and Julia GaoThats good Mojo - Creating a Programming Language for an AI world with Chris LattnerDevelopers are the crown jewels nowGitHub Coplot CLI+WorkIQ+Nightscout...can I correlate which coworker is stressing me out? Satya?Learn to drive stick or the zombies winJohn Maeda failed a Microsoft interview at 17. Decades later he got the jobEPISODE 30 - Scott & Mark Learn To... Get Ready for Ignite
Scott Hanselman |

Scott & Mark Learn To...Fool's Gold

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER