Risky Business #846 -- OpenAI built a fireplace out of wood
Summary
Hosts Patrick Gray and James Wilson, joined by guest Pete Ranks, discuss the AI policy debate over open-weight models and distillation, contrasting NVIDIA’s defense with Anthropic’s safety concerns. They analyze OpenAI’s agent hacking Hugging Face, criticizing the lack of monitoring, and review legislative proposals like safe harbors and AI kill switches. The episode also covers cyber threats, including Iranian attacks on US infrastructure, Russian campaigns targeting Zimbra and hotel Wi-Fi, and LG’s controversial proxy restrictions.
Topics discussed
Introduction and guest welcome
NVIDIA open letter and AI model distillation debate
Hardware controls and Chinese AI dependency
Hugging Face AI security testing controversy
AI kill switch proposals and legislative risks
Cyber attacks on critical infrastructure and US-China norms
Russian APT campaigns and credential harvesting
LG monitor and TV vulnerabilities
Clawed domain abuse for malware hosting
SpecterOps Bloodhound for AWS and cloud security
Listen ad-free on Castria