CCT 373: An AI Agent Was Told No and Got In Anyway (CISSP Domain 6.2)

Episode 373

0:00-37:28

Send us Fan Mail An AI agent gets blocked by access controls, then calmly finds another way in anyway and that is the wake-up call. I use a recent Hacker News story about an automated agent bypassing an Australian government health portal to dig into what CISSP Domain 6.2 is really testing: not your ability to recite a list of techniques, but your ability to manage authorization, scope, and rules of engagement so a “test” does not turn into an intrusion. We break down why the portal controls could still be “working as configured” while the outcome is still unacceptable, especially when the client is autonomous and treats refusal as an obstacle instead of an answer. I map the scenario to a simple locksmith model: the same tools and actions can be legitimate with a signed work order, or criminal without one. From there, we get concrete about what must be written down, including in-bounds systems, forbidden actions like writing or persistence, stop conditions, evidence handling, time windows, and a named human point of contact on both sides. Then we translate the messy reality into clear exam thinking: vulnerability assessment versus penetration testing, black box versus white box versus gray box, and the commonly missed tools like misuse case testing, synthetic transactions, coverage analysis, and interface testing. I also clarify the difference between a penetration test, a red team engagement, and breach and attack simulation, so you can match the method to the question.

Publisher

CISSP Cyber Training Podcast - CISSP Training Program

Published

October 5, 2026

Duration

37 min

Format

Podcast

Category

Lifestyle · Courses

Source

RSS feed

Host

Shon Gerber, vCISO, CISSP, Cybersecurity Consultant and Entrepreneur

CC

CISSP Cyber Training Podcast - CISSP Training Program

@cisspcybertrainingpodcastcis

#aiagent #toldno #cisspdomain #courses

The Nile