SN 1086: The Apex Agentic Adversary - Visual Prompt Injection Strikes
Summary
Steve Gibson and Leo Laporte discuss the restrictive guardrails on Anthropic’s Claude 5, Opera’s new Paste Protect feature against ClickFix attacks, and active exploitation of the Microsoft Bluehammer vulnerability. They also cover Vint Cerf’s retirement from Google, Chrome’s 150th update, and the challenges of controlling powerful AI models. The episode highlights the tension between AI capability and safety, urging better control mechanisms for commercial providers.
Topics discussed
Introduction and show overview
Agenda: AI, Opera, and Vint Cerf
Show notes and Picture of the Week intro
Picture of the Week: Empty street sign
AI model pricing, usage limits, and guardrails
Opera's Paste Protect and clipboard security
BlueKeep vulnerability and CISA warnings
Vint Cerf retirement and internet history
Chrome 150 release and 433 security fixes
EU antitrust fine against Google and chat control
AirDrop and QuickShare security vulnerabilities
Hydration break and personal update
Inkject: Visual prompt injection in AI models
Spinrite 6.3 and SSD performance repair
FATFS vulnerabilities and the apex agentic adversary
AI-assisted coding and future of security
Closing remarks and show resources
Listen ad-free on Castria