Down the Security Rabbithole Podcast

A handful of genuine operational insights emerge (trust = accuracy + consistency + transparency; AI best at focused-but-flexible triage problems; AI poor at novel discovery like finding new persistence mechanisms), but they are buried under movie tangents, broad agreement loops, and repeated 'I think' hedging that consumes most of the runtime.

“trust is accuracy. Yes, but it's also consistency and then transparency. Right. So do you, not only do you get the right answer, but do, can you reproduce that right answer? And then are you telling me how you came to that answer?”

“it's going to be best at like focused problems but not like, but, but, but areas where you want flexibility... a triage is fairly deterministic, but you can't put it in a playbook like soar”

The three-part trust framework (accuracy + consistency + transparency) is a modestly fresh reframe of a common debate, but nearly everything else - SOAR never delivered, AI won't fully replace humans, 99% accuracy claims are marketing - is standard 2024 AI-security discourse recycled without new angles.

“the marketing promises never really quite delivered”

“if your AI stock vendor is saying they're 99% accurate on anything and 100% of alerts, you can give it, like, they're. They're feeding it to you”

Seth has genuinely elite operational credentials - a decade at NSA, joined Mandiant at ~50 people, ran security support for Facebook's business-unit CISOs - making him a real practitioner, not a thought-leader; the episode only partially taps that depth due to its conversational sprawl.

“I spent about a decade at the nsa. I spent roughly another decade at Mandiant... Joined fairly early in that, maybe 50 people or so when I joined”

“I spent a, a few years at Facebook supporting what are roughly their business unit level CISOs”

Almost no concrete data, named clients, or measurable outcomes are presented; the '5-10x per person' productivity claim is asserted without any supporting evidence, and the most specific detail in the episode is Mandiant's headcount when Seth joined.

“now they can do five times or ten times as much per person that's there”

“maybe 50 people or so when I joined”

The hosts occasionally land substantive questions (where exactly is AI best/worst in the SOC lifecycle; will lower tiers be eliminated; will insourcing resurge) but consistently fail to follow up when answers stay vague, and the episode is derailed twice by film references and long host monologues.

“what are the specific, where does it, where is it is this type of technology best? And then at what tasks is it the best at?”

“I just watched the most insane movie the other night. Uh, total sidebar. But, like, it's called Balls up with Mark Wahlberg”