Epizódjegyzetek
This week on Software Sundays, KD breaks down a cybersecurity experiment that shows just how unpredictable AI agents can become when they are given a goal and enough room to operate. We start with an OpenAI cybersecurity evaluation where a sandboxed AI agent found a path outside of its restricted environment, chained together vulnerabilities across multiple systems, discovered zero-days, and even found ways to collaborate with other agents. KD unpacks what happened, why traditional sandboxin...

