
Anthropic Pentagon Rift and 500 Zero-Days [Model Behavior]
Today's episode analyzes the high-stakes friction between the U.S. Department of War and Anthropic, driven by a relationship review following the military's capture of Nicolás Maduro. We detail Anthropic's discovery of systematic distillation attacks by Chinese AI labs using over 24,000 fake accounts to mimic Claude's agentic reasoning. Our technical segment explores Claude Code Security's debut, which revealed 500 zero-day vulnerabilities in production software, shifting the industry from pattern-matching to hypothesis-based defense. We also examine Google's strategic warning for AI startups…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.