This Week in AI

Welcome to the very first episode of This Week in AI! Host Oliver Korzen kicks things off with an exclusive interview featuring Aaron Walls (CEO) and Andrius Useckas (CTO), the founders of ZioSec. Together, they dive deep into the world of OpenClaw security—exposing the vulnerabilities they've discovered in the wild, breaking down real attack vectors, and revealing the essential steps you need to take to protect your agent from exploitation.🔒 Key Takeaways:1. Why Mac Mini might be your safest local setup2. The "Lethal Trifecta" that makes AI agents vulnerable3. How indirect prompt injections hijack your agent through email & documents4. Why 26% of OpenClaw skills contain vulnerabilities5. Input sanitation & judge models as defensive layersZioSec: AI agent assisting in penetration testing | https://ziosec.comTimestamps:00:00 - Why Mac Mini isn't the right setup for everyone12:30 - The right way to host OpenClaw (hint: not AWS)16:11 - Understanding your attack surface27:33 - Why third-party skills are a security risk36:30 - Start with sandboxes & read-only permissions40:03 - Career advice for college students

Google released Gemini 3.0 this week, and I’ve been hands-on with it every day since launch. After diving into all the new capabilities, performance upgrades, and quality-of-life improvements, I’ve distilled the most important updates you should know. In this breakdown, I walk through the biggest feature changes, what they mean in real-world use, and how Gemini 3.0 stacks up against the other frontier models.