Claude Was Told “No Internet”: It Hacked 3 Companies Anyway
In this episode of AI & Cybersecurity, Leor and ToxSec break down what happened when Anthropic’s cyber evaluations reached real-world systems, why role-confusion attacks can make AI mistake malicious instructions for its own reasoning, and the growing fight over whether frontier AI needs a global “brake pedal.” We also dig into the EU’s new AI […]
Continue Reading