SN 1086: The Apex Agentic Adversary - Visual Prompt Injection Strikes
Summary
Steve Gibson and Leo Laporte discuss the strict new safety guardrails on Anthropic’s Claude 5, which hinder security research, and Opera’s new Paste Protect feature against clickfix attacks. They also cover the active exploitation of the Blue Hammer Windows vulnerability, Vint Cerf’s retirement from Google, and the ongoing debate over AI control versus capability.
Topics discussed
Intro and preview of topics
Listener picture of the week and sponsor intro
Anthropic's Claude 3.5 Sonnet controversy and fallbacks
Opera browser introduces Paste Protect feature
Bitwarden password manager and digital legacy
Microsoft Blue Hammer vulnerability and CISA response
Vint Cerf's retirement and legacy of TCP/IP
Sponsor: Exbow agentic pen testing
Chrome 130 update with 433 security fixes
EU fines Google and Apple over antitrust and encryption
Vulnerabilities in AirDrop and Quick Share protocols
Inkject: Visual prompt injection attacks on VLMs
Spinrite 6.1 SSD repair and optimization
Sponsor: Adaptive AI security
Critical vulnerabilities in FATFS file system library
Outro, GRC.com resources, and Club Twit
Listen ad-free on Castria