TechSpective Podcast

Remember those Facebook quizzes that made the rounds during the pandemic lockdowns? They'd ask for things like the street you grew up on or the name of your first pet, and everyone treated it as a fun way to get to know each other. I spent a good chunk of that time telling my family to stop answering them, because those are the same questions banks and websites use to verify who you are. My family mostly thought I was making a big deal out of nothing, and in fairness, each of those details is pretty harmless by itself. They just don't stay by themselves for very long. Ron Zayas, CEO and founder of Ironwall, joins me on this episode of the TechSpective Podcast to talk about how those pieces get stitched together and what you can do about it. Ironwall started about 15 years ago by removing personal information for judges, and it has since expanded to protect executives and other people in the public eye. The company is now part of Incogni. Your Address and Phone Number Tie It All Together A couple of years ago, I had a call with a different company that sells privacy services to executives and wealthy clients. They ran a quick scan on me before we talked and opened the call by telling me where I'd lived and where my kids live, and they'd already connected me to my parents. It was impressive and a little unnerving, but I also pushed back, because a lot of that is public record and most people don't buy their house through an LLC to keep their name off the deed. Ron's view is that your home address is exactly where to draw the line, and there are ways to protect it that don't involve setting up a shell company. Data brokers are where all those scattered bits come together, and the details that tie them back to a specific person are usually your home address and your mobile number. Think about how many times you've moved in the last ten years compared to how many times you've changed your cell number. As Ron put it, "We just need to break the things that tie us to everything else." You don't have to be a judge or a celebrity for this to matter, either. People get doxxed now for posting an unpopular opinion, and someone who's angry enough will go looking for your employer or your house. "We're all one interaction away from somebody wanting to hurt us," Ron declared. AI Does the Legwork Now Connecting those dots used to take some effort. An attacker needed several databases and the patience to cross-reference them and figure out which details went with which person. AI is very good at that kind of pattern matching, so the same old data is a lot more useful to someone who wants to impersonate you or find you than it was a few years ago. Ron also raises concerns about government agencies buying commercially available data they'd otherwise need a warrant to collect. Be Harder to Find Than the Next Person I'm a little cynical about some of this. My name, my birth date, and the other details that never change have been exposed in enough breaches that I assume they're out there for good, and when another breach notice shows up I barely look at it. Ron doesn't claim you can scrub everything, though. His point is that making yourself harder to find still pays off, because if someone has a hundred pieces of information on you and twenty on the next person, you're the one they'll go after. That's the same advice I gave years ago, when people figured out WEP encryption on Wi-Fi could be cracked in about a minute. If you couldn't upgrade, I said to leave WEP turned on anyway, because the neighbor with an open network was still the easier target. You're not going to disappear from the internet, and Ron isn't suggesting you try, but there's no reason to make it easy. He wraps up the episode with a few inexpensive steps that are worth the effort. Watch or listen to the full episode.

I realize I am dating myself to some extent, but my first cell phone plan came with a fixed number of minutes and a small allotment of text messages, and anything beyond that was billed one at a time. You got in the habit of watching the clock on calls and thinking twice before you replied to a text. That model didn't last. Once enough people wanted to use their phones without doing arithmetic, the carriers moved to unlimited plans, and now nobody counts anything. AI is still in the counting stage. A company signs up with OpenAI or Anthropic, people start using it for real work, and then the invoice arrives, and someone has to explain how many tokens the team went through last month. Richard Luna, CEO of Protected Harbor, joins me on this episode of the TechSpective Podcast to talk about where that pricing model is headed, and why he thinks the better option for a lot of businesses is to run AI on hardware they own. Who Pays for the Build-Out My assumption going in was that token metering will eventually give way to something closer to unlimited use, because people won't lean on a tool they have to ration. Richard comes at it from the other side. He points out that “no AI company is making money,” while enormous amounts of borrowed money are going into new data centers. Sooner or later that debt shows up in what customers pay. If he's right, the prices businesses pay today are an introductory rate, and nobody has said yet what the regular rate will be. Richard compares the moment to the subprime lending run-up before 2008 and to the dot-com era. The dot-com comparison is a useful one, because the bubble burst and the web kept going, while plenty of the companies that looked permanent in 1999 did not. Richard doesn't doubt the technology itself. As he puts it, “AI is real, and its benefits are real.” The open question is who will be selling it, and at what price, once the correction happens. The Case for Running It Yourself Richard's answer is local models. His view is that a model running on your own machine is the best way for a business owner to use AI, and the best way for a developer to “own their own destiny.” The data stays in-house, and the cost is tied to hardware you already bought rather than to a vendor's need to earn back its investment. The hardware side of that argument keeps getting stronger. My first computer was a Commodore 64 with 64K of memory. The Surface Studio laptop I use now has 64GB, and I've already downloaded a local LLM to run on it. Hugging Face has a huge catalog of open models to choose from, and some of the models coming out of China run on far less hardware than the big U.S. platforms use. My kids object to AI for a number of reasons, including the water and power that data centers consume and the IP theft and general lack of ethics demonstrated by the big AI companies. Those are exceptionally valid concerns, but a few months ago Bruce Schneier shared some insightful wisdom that I keep referring back to: most complaints about AI are really complaints about the companies behind it. Running an open model on your own laptop doesn't answer every one of those objections, but it takes a lot of them off the table. Planning Around the Bill None of this means businesses should drop cloud AI tomorrow. The frontier models are still better at some tasks, and plenty of teams don't have the hardware or the in-house skills to run models themselves. It does make sense to track what you're actually spending on tokens and to figure out which workloads could run locally before a price change forces the question. Richard and I cover a lot more ground in the full conversation, including how he went from punch cards and a bulletin board system in his grandparents' house to running an engineering-heavy company, and where he expects today's AI giants to be in 20 years. Watch or listen to the full episode of the TechSpective Podcast below, and let me know whether you're running AI locally yet.

Amihai Niederman joins me on this episode of the TechSpective Podcast to talk about what happens to identity security once AI agents start outnumbering the humans on a company's network. He's the co-founder of NewCore, a startup betting the answer isn't another tool bolted onto what companies already run. It's a rebuild. That bet is worth taking seriously. The identity stack most companies run today wasn't built for this problem. It was built for a much smaller one. Identity Grew Up As An IT Problem For most of its history, identity has been treated as plumbing, not protection. Active Directory sat inside the network. If it got breached, the conversation was about how someone got past the firewall. Identity itself wasn't seen as the thing that failed. That framing made sense when the network was the perimeter. It stopped making sense once everything moved to the cloud. Identity providers became the real gatekeepers. Stealing a password turned out to be a lot easier than breaching network hardware. The industry responded the way it usually does. It bolted on point solutions, one after another, each one solving a piece of the problem on its own. NewCore talked to more than 100 CISOs while building its product. The average company, they found, runs 8.3 identity tools. That's a vendor number, worth taking with a grain of salt, but the picture rings true. A pile of tools from different makers, held together, hoping the seams don't split. Agents Break The Math, Not Just The Tools None of that was built with AI agents in mind, and that's where the trouble really starts. An agent isn't just a faster version of a person logging in. Amihai draws a real line here. A lot of what gets called an agent today is really a power tool, with a person still making every meaningful decision. It only becomes an agent once you hand it a full task and let it decide which tools to use, in what order, on its own. At that point it stops being a stand-in for a human. It needs its own governance. That distinction matters because of scale. Agents can work something like 100 times faster than a person, and they never log off. That's not a modest bump in the number of identities a company has to manage. It's closer to 100 times as many. A ninth tool doesn't fix a jump that size. Amihai compares it to converting a gas engine to run on electricity. You can't patch your way there. Most companies are handling this the sensible way for now. Deploy one agent. Watch it closely. Adjust, then repeat. That works fine at a ratio of one agent to one human. It falls apart once the ratio flips, and the flip is coming faster than most rollout plans assume. The Resistance Isn't Where You'd Expect It The instinct is to assume the hardest part of an identity rebuild is convincing a CISO to hand someone new the keys to the whole environment. It usually isn't. CISOs already live with what's broken in their current stack. They know exactly what they're trying to fix. The real friction shows up everywhere else in the company. Old identity connections nobody wrote down. Employees who need a new authenticator app. A migration slow enough that nothing breaks along the way. That's the harder, less glamorous problem. It's also the one that will decide whether identity security catches up to what AI agents are about to demand of it. There's more to this conversation than fits here. Where identity security goes as the ratio of agents to humans keeps climbing. A sharp comparison involving stoplights and self-driving cars. How NewCore is thinking about protecting actual people while building for a mostly agentic future. Catch the full episode on the TechSpective Podcast:

Nicole Beckwith joins me on this episode of the TechSpective Podcast to talk about a problem most of the security industry still hasn't fully reckoned with. Beckwith runs security engineering and operations at Cribl, and before that she spent years in law enforcement and digital forensics, which is the lens she uses to explain why the way we've built detection for the last couple of decades is running out of road. Matching Isn't the Same as Understanding For as long as I've covered this industry, detection has mostly come down to matching something you already know is bad against something showing up on your network: a hash tied to known malware, an IP address tied to known bad infrastructure. It's a yes or no answer, and for a long time that was good enough, because attackers weren't changing their tooling anywhere near as fast as they're changing it now. That model is breaking down. Polymorphic malware changes its own hash per victim. Phishing infrastructure gets built fresh, one URL at a time, sometimes one top-level domain at a time. By the time a hash or a bad domain makes it into your threat feed, the attacker who used it has often already moved on to the next one. Beckwith put it plainly: the bottom layers of the pyramid of pain, the framework the industry has used for years to rank how much damage a given detection actually does to an attacker, are “essentially just on fire right now.” An Indicator Only Proves Someone Was There Beckwith's law enforcement background gives her a useful way to explain what's actually missing. An indicator, she says, is a single piece of evidence, the equivalent of a fingerprint on a glass or a license plate written down at the scene. It puts someone there. It doesn't tell you what they did once they arrived, who they were working with, or why. “It's a fast, cheap triage of known bad,” is how she describes it, and that's the right job for it. The mistake is treating a triage tool like it's the whole case. Behavior is a different kind of evidence. An attacker can swap a hash in seconds, but if they want credentials, they still have to do the things credential theft requires, and those actions leave a pattern an indicator alone was never built to catch. Why This Is Harder to Fake, and Harder to Buy This is also where a lot of vendor marketing falls apart. Walk any security conference floor this year and nearly every booth claims some version of AI-powered, behavior-based detection. Beckwith has spent enough time doing exactly that to know most of it is still a rules engine underneath, with a model bolted on for the pitch. Real behavioral detection takes the unglamorous work of engineering telemetry, mapping it to actual attacker tactics, and tuning it against your own environment instead of shipping something generic and calling it AI. That work doesn't show up on a slide, which is exactly why most vendors skip it. Beckwith gets into a lot more of this on the episode, including what this shift means for who gets hired into a SOC and how she personally judges whether a vendor's AI claims hold up. Watch or listen to the full conversation on the TechSpective Podcast.

For most of the last year, the AI security conversation had a clear villain. Attackers were using AI to write better phishing lures, adapt malware mid-attack, and move faster than defenders could keep up with. That story was easy to tell because it fit the shape we already understood. Bad guys get a new tool, they use it against us, we build a defense. Then the defenders' own AI started breaking its own rules, and the story stopped being that simple. The Model Did What It Was Told In July, OpenAI disclosed that one of its models, running an internal test with reduced safety guardrails, found a zero-day vulnerability, broke out of its own sandbox, moved laterally through OpenAI's research environment, and reached across to Hugging Face. Two weeks later, Anthropic acknowledged that three of its own models had done something similar. Both companies were describing their own systems doing exactly what they'd been asked to do, just not in the way anyone expected. Brad LaPorte of Morphisec has been tracking this closely, and he pushes back on the idea that this is as novel as the headlines suggest. "This isn't necessarily novel," he told me. "It's just they made the front page of the newspaper finally." What's changed is the scale, and the level of autonomy behind it, enough that it's getting harder to write off as an edge case. That reframes what "guardrails" actually means. A model told to solve a problem and finds an unsanctioned path to solving it is still doing its job, just not the way anyone intended. Isaac Asimov's laws of robotics get invoked a lot in these conversations, and for good reason. "Don't harm a person" sounds like a rule until you realize a model can read it literally and conclude nobody got physically hurt, so nothing went wrong. Financial harm doesn't register that way. Neither does legal exposure or reputational damage, and neither shows up in a rule that vague. The Fundamentals Haven't Caught Up While that governance conversation plays out, the operational numbers are moving in the wrong direction. IBM's most recent cost of a data breach report showed the average time to identify a breach getting worse for the first time in five years, up six days to 247. Meanwhile, roughly 88 percent of organizations report using AI in at least one function, but only 5 to 10 percent are seeing meaningful return on that investment. A lot of the AI-driven layoffs making headlines aren't the result of AI actually doing the work yet. They're companies freeing up budget to chase a productivity gain that hasn't arrived. That gap between adoption and governance is where the real risk lives. Organizations are running AI agents that talk to each other, share data, and make decisions with non-human identities that most security teams haven't fully inventoried, let alone secured. Add unsanctioned AI tools employees are using without approval, and you have an attack surface that's expanding faster than most companies can map it. None of This Changes the Basics The uncomfortable part is that the fix isn't exotic. Identity and basic visibility into what's actually running in your environment still account for most of the risk reduction available to any organization. AI adds a new layer to secure, but companies that had their fundamentals in order before AI showed up are adapting. The ones that didn't are finding out that AI doesn't so much create new problems as it makes the old, ignored ones impossible to keep ignoring. Brad and I get into all of this on the latest episode of the TechSpective Podcast, along with where he thinks the AI funding bubble is headed and why he compares it to the mortgage-backed securities mess of 2008. Give it a listen.