Gemini 4 enters the ring, MI5 flags research ties, Custom GPTs deliver malware
Summary
Sarah Lane covers Google's Gemini 4 Argon release, MI5's warning on Chinese research funding, and a ClickFix attack via malicious custom GPTs. The episode also details a Cisco SD-WAN zero-day, Pentagon orders for cyber operator wellness, and deceptive behaviors in Chinese AI agents. Finally, it highlights Google DeepMind's new watermarking system for AI-designed proteins.
Topics discussed
Introduction and show overview
Google Gemini 4 Argon launch and benchmarks
Google DeepMind internal results and employee skepticism
MI5 flags Chinese research funding and UK academic links
Custom GPTs used for ClickFix malware attacks
Cisco patches actively exploited SD-WAN 0-day
Cisco SD-WAN vulnerability details and mitigation
Sponsor segment: Intezer AI SOC
Alex Stamos joins Cognition as Chief Security Officer
Pentagon orders Cyber Command to improve operator support
Details on Cyber Command wellness and recovery measures
Chinese AI agents show signs of deception in studies
Google DeepMind develops watermarks for AI-designed proteins
SynthID Bio system mechanics and testing results
Limitations of protein watermarking and security keys
Outro and feedback information
Listen ad-free on Castria