Security You Should Know

There are a handful of genuine ideas - the accountability vacuum when AI makes errors, why training proprietary models is wasteful, and the atomic-vs-investigative speed trade-off - but they are diluted by repeated agreement-piling, generic AI enthusiasm, and conversational filler. The ratio of novel claims to airtime is low.

“AI brings the infinite problem to both sides of the equation, right? As long as whatever tool you're using to speed up your own efficiency is able to be used by the attackers, it ends up being kind of an infinite game.”

“the reason Atomic Works is it's faster than any investigator. Right? Like knowing this is bad, turn it off in 10 seconds, you're not going to have an investigator that's faster than that.”

The analyst-accountability argument - 'I get all of the risk, none of the reward' when AI errs - is a crisp, underappreciated framing. Everything else (AI noise, data sharing hesitancy, community threat intel) is well-worn SOC discourse recycled without a genuinely fresh angle.

“If AI makes a mistake and I still get blamed for it, I get all of the risk, none of the reward.”

“we're moving from an indicator based organization where that sort of threat intel sharing becomes really meaningful because there's an atomic indicator that has a short TTL”

Howard Holton is a GigaOm analyst - a thought-leader role rather than an operator who has run a SOC at scale. Will Gregorian is a practitioner CISO but at a single medical organization of unspecified size. Grant is a co-founder pitching his own product in what is explicitly a sponsored showcase, limiting independent credibility.

“Helping us find out why this is such a problem are Will Gregorian, the CISO over at Galileo Medical, and Howard Holton, the CEO at gigaom.”

“All companies showcased on this program are sponsors of CISO series.”

Almost no hard data appears: no alert-volume reduction figures, no customer outcome metrics, no case studies, and no named customer references. The only concrete specifics are brief tool-name drops (CrowdStrike, Wiz) and a rough customer-size threshold of 'a couple thousand people or more,' which is too vague to be actionable.

“core customers tend to be in this day and age we're talking about analyst overload... organizations of a couple thousand people or more”

“you give a uh, problem to Claude on security analysis, it's a coin flip whether you're going to get the right answer or not”

Howard's sustained, pointed pushback on data-sharing inertia and AI accountability creates genuine tension and produces the episode's most substantive exchange. However, the host operates from a rigid 'three essential questions' promotional template and never challenges a single product claim, keeping the ceiling low.

“We've got to figure out how to get over this. Security through obscurity is almost what it is”

“I'll be honest, I do not like it. I understand why it happens, but I do not like the response.”