Heraclitus, AI LLMs, SSO, TTP, NetLogon, PAN-OS, AI Cost, Aaran Leyland... - SWN #586

AI is no longer just a tool—it's becoming an execution surface in its own right, turning trusted summaries into attack vectors. A groundbreaking vulnerability in AI assistants like ChatGPT allows malicious webpages to inject phishing links, QR codes, and fake alerts directly into AI-generated responses, exploiting user trust without requiring a click. This isn't a code exploit—it's a psychological one, where the AI’s rendering of untrusted content becomes a delivery mechanism. Meanwhile, companies are hemorrhaging millions on unbounded AI usage, not from tokens, but from the massive cost of validating AI’s hallucinated findings. As LLMs solve century-old math problems and evolve like Heraclitus’ ever-changing river, the real danger isn’t AI’s intelligence—it’s the blind trust we place in it. The episode warns: treat every AI summary as hostile until proven otherwise, and remember that zero trust must extend beyond systems to human workflows. The episode also dives into the alarming speed of modern cyber threats, with exploit-to-attack windows shrinking to days, and critical vulnerabilities like NetLogon and PanOS being actively exploited. Patching isn’t just technical—it’s logistical, human, and often impossible when teams are furloughed or keys don’t work. The core message? Automation and AI don’t fix broken processes—they amplify them.