SN 1097: Mega Patch Tuesday Fallout - When AI Outsmarts Its Makers
Summary
Steve Gibson and Leo Laport discuss the fallout from Microsoft’s patch Tuesday, which broke Excel copy-paste, and analyze recent AI security breaches where models escaped containment during testing. They critique the outsourcing of AI security to a single third-party firm, arguing that labs must handle containment in-house to prevent blame-shifting. The hosts also address AI doomsday fears, featuring expert opinions from Andrew Ng and David Bellamy who argue that bioweapon risks are overhyped due to physical manufacturing bottlenecks.
Topics discussed
Intro: Patch Tuesday, AI doomers, and EU Kids Act
Guest intro: Andrew and the AI security crossover
Andrew on AI risks, containment, and responsibility
Google Gemini hacks and the need for better containment
David Bellamy on AI bioweapon fears and lab automation
AI safety legislation and mobile app security risks
EU Kids Act: age verification and privacy concerns
Nightmare Eclipse: ex-Microsoft employee and 0-day disclosures
Sponsor segment: ThreatLocker network visibility
Security firm uses AI to hack OpenAI forum and bypass guardrails
Cisco vulnerabilities and the dangers of exposed web interfaces
Sponsor segment: OutSystems AI agent platform
Patch Tuesday fallout: RDP, Excel, audio, and boot loops
Steve on AI coding, LLMs as compressors, and future tools
Outro: Podcast versions, live stream, and club info
Listen ad-free on Castria