AI Can Emulate Any User Behavior in 5 Mins: Warns STIX/TAXII Architect

AI is nuclear level technology. You cannot just give anyone unlimited access.

Where are we today in terms of our capability to thwart those kind of efforts.

So you've given super intelligence to individuals that are not operating at the same governance level. Now the world is ungovernable. This ability to deploy the pieces that escalate authority in a way that's so fast. By the time I build a framework or update my MITRE ATT&CK control slayer to understand and simulate the attack and update the rest of the world, there will be 100 new techniques put into place by the time I release. When you turn a nonprofit organization into a for-profit entity that was designed to do nonprofit research in protecting and securing and governing and making AI safe, you clearly have sent a signal that that's not relevant.

All right, Ali Khan, welcome to the pathway to peak performance. So great to have you in. I'm so excited for this episode. Thank you for coming.

Thank you. Thanks for having Yeah.

Uh my pleasure. So, uh your charity is UNHCR.

Great. They do a lot of great work globally. When there was a earthquake in Pakistan 20 years ago, I remember volunteering with them. So, I feel like they've done a lot of great work after natural disasters.

Phenomenal. Uh, and all over the world they're they're operating and doing great things. Okay. The show is sponsored by KetoneAid, which is like the true V ketone. Uh it's about 10 years ago the actual research came out um on this and it's really uh the Ferrari of ketones. This is uh really amazing stuff. So what you can see here is we we haven't popped the top on it. So I'm going to give this to you and I'm going to let you put the uh capfuls in the water and we're going to take those and then see how we do. Bottoms up.

That's right. Cheers.

Interesting. Unique.

Yes. Very unique.

Okay. So, what we're going to do, uh, it's 2:57, so we're going to come back in about 20 30 minutes. Okay. Check in to see if we feel a little, uh, extra brain power.

Nice. Let's do it.

Cool. So, uh, first things first, let's get into the origin story on STIX uh, TAXII. Uh you have an incredible CV. Take us through it.

Yeah, thanks for having me. I think this is a great way to start what I really stumbled upon early in my career. I had a mentor, Mark Clancy, who is the chief security officer at T-Mobile right now at the time. He was the chief information security officer of Depository Trust Clearing Corporation. And so he told me he was working on a pretty interesting project. He thought I'd be a great product manager lead for it. And it was a bit of a startup that we were building uh never been done before.

It was the structured threat information expression language. It was a protocol to share cyber threat information real time from machine to machine. And as you could see today with MCP servers and a lot of autonomous agents transacting information this is becoming more and more relevant. So machine-to-machine communication in cyber threat information was something foundationally no one had architected. When we talk about all the different layers web application firewall, network firewall, different layers of application threats as well as hardware threats as well as actual threats on the browser on the endpoint. So we piecemealed everything.

We brought together experts for MITRE, Department of Defense, Chief Information Security Officers from different banks, the government, uh, Department of Homeland Security, Science and Technology Team. At the time, it was Richard Struse who was leading it. I think it was then later rebranded to CISA cyber security infrastructure security agency which has had a lot of news unfortunately uh recently but when Jen Easterly took it over uh a lot of this work had already been operationalized so um I think I was at the right place at the right time in my career.

I had not worked in cyber security and I was really excited to work in it. I had worked in a lot of product management and product builds and my mentor said don't worry you don't need to be an expert in cyber security you just need to make sure this team goes at the right direction so I started my career in cyber at that time and the first thing you imagine yourself to start your career in an industry at such an important juncture at a such an important project which I had no idea We had a top secret name for the project called Avalanche. So we didn't even have a name for the startup. We didn't have a name for STIX TAXII, you know, it was all early versions, right?

So then uh people started stumbling into the room as things evolved from Washington DC meetings in Tampa, Florida in different locations where we would meet up people from MITRE like Mark Davidson was a great peer of mine who's uh my coworker at the startup that we built called Soltra as well as one of the pioneers in the MITRE ATT&CK framework early in the days of STIX TAXII. We had Richard Struse, you know, great gentleman who went on to build other cyber security startups later in his career and uh contributed heavily to MITRE ingenuity.

But um to answer your question, essentially think of it like this. If you're trying to build a factory and you want to send a million bottles of Coca-Cola, you probably want to have a standard way to package, ship, and deploy, not just putting it in the bottle, tightening the bottle, shipping it into the truck, and to the point of retail locations. So it's the simplest way of looking at how do you take information and then transport it where everyone understands how to ingest it on their side.

That was the political side of the problem where you had vendors at the time McAfee Symantec right and not CrowdStrike or Palo Alto Networks like we have today right at the time you had different vendors who had an interest in the game. Then you had all the large banks that wanted to have an influence on how the protocol would be designed. Then you had the government side and then you had the actual startup which was us building the protocol as we were operationalizing it. So it was a very challenging first job in cyber to take government public private sector and get them to agree on something.

You know, I'll give you an example. We were debating for 3 months, right?

Yeah, it's still a problem today.

We were debating whether this should be in XML or JSON XT, which was a huge debate, intellectual debate. People from Blue Coat, you know, were involved uh and which is a great company acquired later on in network security. And you had a lot of really good reasons why we could go one direction or the other. We had I would say geniuses in the room like Silu from who's the chief cyber security scientist at Bank of America. Patrick Garity who contributed to vulnerability research you know and MITRE Bret Jordan you know was a pioneer in network security protocol architecture Mark Davidson right Aaron Shernan Mark Clancy you know like and I was in a room full of these individuals you know as a firsttime product lead on a top secret project essentially meant to help even NATO operationalize cyber threat information today right and working back into that Um, foundationally, you know, I felt like, you know, as a product manager, you keep things simple. You get there every day, you show up, get the team to agree on requirements, and you start engineering and testing and start deploying and start working with customers.

Luckily, our subject matter experts were some of the top cyber security detection engineers working at some of the largest banks in the world, giving us this feedback. So stumbling upon this was like a fascinating project that I learned so much right as a firsttime product manager in cyber security and for me the STIX TAXII protocol was essentially the first attempt in the cyber security industry uh not just in the US but globally to create a standard mechanism for machines to share information as Google engineers are doing today with the MCP server for agents.

So um we we had hundreds of vendors many years later come on board and use the protocol to share information. We had multiple banks that were not only leading by operationalizing it but then hundreds and thousands of banks that join later to share that information and ingest it. And uh I felt like the mission of the startup wasn't to really just go from series A to B to C. We were really designed to really advocate this protocol across the globe to uh create awareness around the importance of sharing cyber threat information. So it's a it's not required you know and then we were able to convince that we should start sharing this information and and made a really good headway as well as a result of it. So there's so many ways to skin this cat right. So, so that was my uh experience uh from day one building the protocol.

That's such a cool um story to think just like hey you you know new guy first time in you walk into the room and it what what a great what the best thing you could ever have happen is walk into a room where everybody has tons more experience. All you got to do is just show up on time make sure the project goes the right way and listen to everybody everything everyone's saying. Um, and along the way you pick up so much from so many great people. How cool is that?

I think that's why they picked me. They're like, "This guy has no skin in the game. He has no bias. He's very mature and very methodical and very efficient. He's going to bring people together. He's not going to pick side. He's not going to try to play favorites because he's written a thesis on why JSON XD is the best way to operationalize, you know, lightweight information into computers, right? Versus XML. meant he doesn't work for Microsoft so it's not going to be you know so there was a lot of baggage I think my peers that they bought so much experience that were involved in making this decision that I didn't have so I could really focus on how do we get this in the hands of people so we stop cyber attacks

I mean that's it's a really brilliant decision on on their part because if you get too attached to any model today any large language model today it's it's actually you know a liability because uh you know if you look back in the way that traditional software was developed. You know, people made their their career being an expert in this one area and then a sword fight with everybody about why their way is better. And um that notion of you coming in and not having any skin in the game is cool because you have this ability to be open to all things and to see maybe some of the blind spots that some of the other folks might not be seeing. And when you're doing something in a public private, it's that's a complicated thing in and of itself.

Yeah.

You've got to be really as uh neutral and and and make it happen. That's super cool.

Yeah. Because the KPIs were different, right? I mean, in a private startup, you're looking to raise capital, hit ARR, and grow, grow, grow.

Yeah.

With the public sector, you want to make sure safety first. Is this going to work in defense networks? Is this going to work? So, hardening, testing, maturing the protocol was so critical to get it to 99.999, right? Uh availability whereas in the private sector it's quite the opposite. You want to get it to fast to market fast to market and then work backward.

Yeah. I mean and when the stakes are so high um it I mean it's it's a critical deployment. Um that's that's a big deal. Uh so what a cool what a cool story. U kind of brings us to last April at NC and CK had an update. Uh there were two uh pieces in there. Um one was uh stealth and the other one was uh deception evasion, right?

Yes.

Um it comes at a critical time.

Yeah.

When 87% there's a recent uh survey of um uh cyber security professionals, I'm sure you're familiar.

Yep.

And 87% of them said that the you know the threat level was dramatically up due to AI. Uh they're able to get in a million times faster, faster and emulate legitimate behavior.

Yeah. Because anomalous detection has been perfected in the cyber security industry from an engineering perspective. Very well researched, peer-reviewed for decades now. If you look at anomalous detection, what is it? It's uh you know, Jock, right? Doing uh something to write in their notebook with a certain calligraphy style, closing the book at a certain speed rate and we can understand that behavior. And so in uh computer terminology, it would be you logging in at a certain period of time and logging out at a certain period of time and we have that data, we've collected it and then if something anom anom anomalous happens, we're able to flag that.

So agents can train on that data that took you and cyber security engineers decades to understand and build logic around and code around uh to reduce cyber attacks and reduce risk. An agent can train on that within 5 minutes. So now you just went from you're right like you know we reduce risk down 90% to we just open the floodgates of risk you know because emulation deception is now much easier not just deep fake but if I compromise your endpoint and I ingest that into a localized large language model and start training on how you log in which browser which websites you visit which is very easy to download from you know hashes and browser or cookie data. It's all sold online, right, as well.

And then I apply that to a large language model within five minutes. That model can tell me how to emulate that behavior and now go and and legit legitimately attack a certain computer from your IP address, which now you know that authentication that door would be, you know, open pretty easily. So it's yeah it's very very dangerous what we're doing uh with AI and how AI governance has been 99% of the investment has gone into AI innovation 1% into AI governance right so we're we're playing way behind right in terms of what we could actually do to catch how bad it is the risk

I mean you think about that um most recently Mike Johnson sent the Senate home the Congress home excuse me um and we have no there's governments governance from now until November right when they go back into session and so there's a lot you know uh California with the kill switch and um there's a lot going on with it and it's coming at a point in time when 47 another survey 47% of um professionals said that they're not ready for it. They're not ready for these level of attacks um and they just are sort of like scratching their heads.

So this is where these standards uh that you've put into place, these 15 tactics that are in place become so so important to people to understand. And if you're behind, that's the first place you should go and try to figure out exactly, okay, what are you doing? How do you get a gap analysis around where are we weak? And then probably engage some somebody like you, you know, um to to actually help them.

Yeah. Operationalize it.

Yeah.

And validate if it's working because um a lot of the security vendors do a great job in keeping up with MITRE ATT&CK framework making sure all the controls and all the updates that are released are actually validated into their own uh agentic infrastructure as well which is great uh but how do you validate them? How do you pentest that? How do you make sure it's actually going to th more sophisticated attacks because software is liquid right? I mean, with every software update release on your iPhone, iPhone, on your iWatch today, as we're sitting here, it's likely going through a release of one application, right, that just rolled out a hot fix.

So that static time period at 3:19 p.m. which was a minute ago to now 3:20 p.m. your software's uh amorphic graph neural network threat model has completely changed within that one minute. It's crazy. So now I can release right a new exploit against that new release which hasn't been patched. It's it's a very cat and mouse cat mouse game. It's a very liquid game. It's not static. And that's what makes cyber security so unique from every other industry.

It's not really an industry. I um I always have this debate with people. I said cyber security is not an industry. It's like infinitely embedded in everything when we talk about technology. It's not an isolated I know it's easier for econ economists to put in that category, but it's really embedded in everything that you do digital.

Yeah. And it's ubiquitous. I mean it's just without it uh we got real problems. I mean and you know when you think about it you take a look at health care right and that's the number one place for attacks um and you know you've got these ransomware so first things first you know if you have a breach of PHI you've got to report it uh immediately uh and to CISA and uh and other agencies the FBI and Treasury does not want you to pay the ransom I mean and it makes sense right I mean don't don't reward those guys anymore um it's already breached Anyway, you're going to pay the penalty and you're going to be on that list uh being watched for whatever number of years it is.

Anyhow, the real risk is that you've got a lot of people that still have these servers that are running in their practices, healthcare. You know, you start with like a small practice and you kind of work your way up into maybe a group practice or a network uh you know, an IPA into maybe a regional network or into a large hospital. Now, Cedars-Sinai and Stanford, they have the infrastructure, right? They're prepared. They have the budget, the

Yeah.

Um, but many of the other folks that are, you know, down the hill from that need to really be thinking about, hey, I've got to get up into the cloud where there's a a security layer that's, you know, working all the time because what you just mentioned about software being liquid. Um you need a a dedicated team that's at that layer uh to be looking at these kind of you know looking at threat detection in a way that there's just no way I mean you know the poor guy that leaves the server and there's some some way to you know get in

Misconfiguration right you know subception attacks data poisoning you know you could do a lot of really cool stuff I don't want to say the word cool but you know as an offensive or white hat hacker right I mean it's pretty cool what you could to to decoy or delay certain systems, misconfigure them and mistime the release of information to start a certain type of attack, right? Especially if you're moving laterally and from one place to another.

I think when we look at health care and we look at our private data unfortunately you know I've been working in this industry for so long and with uh the great folks at the healthcare ISAC for example like Errol Weiss you know um these people they understand the volume of attacks targeted at smaller non-sophisticated low lower cyber security budget institutions is pretty high right because the time to attack to time to entry is much lower from targeting, you know, a much higher cyber security budget, a more sophisticated team. So, and then you could still get the same quality of information, right? If you're trying to shut down their operation, so now they have to pay you around.

So, it's financially motivated for them versus shutting down the operation of a larger um sophisticated cyber security team where they're likely not going to pay the ransom because they're the ones likely leading and and it's a um a risk for them from a reputational perspective. Reputational damage to pay a ransom after you've been compromised is much higher for that institution I think than it is for a law.

So I think the way we solve that is again sharing cyber security information sharing cyber resources in a decentralized mechanism can help us reduce attacks. One great example of that is that if I have a centralized console of information that I'm storing and that has a localized admin account and the attacker is trying to, you know, penetrate my administrative account and hijack the ad administrative account so they get lock me out from the rest of my organizational uh assets.

The way I see that is that if I can work with the top 20 uh healthcare institutes, look at the number of incidents they've had every quarter, get on a quarterly call with them, beg them to share that information with me, understand what the incident was, and now with the power of AI actually distill it down much faster and start to build real time localized models to help me digest top attacks, how they happen, reverse engineer the incident and map it back to the MITRE ATT&CK framework and then see which controls I'm missing and where I might be hit and only pass those I likely reduce 99% of my attacks which is great right

And so information sharing is very powerful it depends on the culture at your organization and the budget unfortunately that you have or the time that you have to stay up to date with right it's why you create content right because you want to stay up to date with the latest that's going known in cyber tech AI right in in healthcare. So it's the same way in the cyber security. If you're not tapped into your consortium and your community and you're not using the power of the community, you're likely spending way too much money on information you could have acquired from someone else's successful incident that could have informed you because it's the same it's likely the same attacker or the group that's going to attack you as well.

So you what people don't understand is the attackers are way smarter than you think uh or most people think and they like doing what they do. They enjoy it like this is something it's a different form of motivation for them.

Absolutely. And so it's a it's a different animal. Um it's interesting you mentioned like you know hey if you could talk to the top you know organizations and they could share that you can reverse engineer makes perfect sense. Perfect sense, right? That idea of information sharing. You look at Ascension, they're shut down for 18 days.

Yeah.

Right. Where they can't operate at all. You got docs writing on paper charts and they did get in the admin panel. They did shut everybody out and there's there's no, you know, there's nothing you can do at that point in time. The backups at that point are

Yes. Compromise.

That's the interesting thing too that I'd like to understand more about in this is how this how AI is helping to escalate that ability once you're in. So you know like there's that one reports like what 29 minutes to get in and get a foothold then that the fastest was 20 that you know 27 seconds somewhere in between there right there's like this ability and then this ability to deploy the pieces that you've talked about and then escalate authority in a way that's so fast it's a it's incredible. I'd love to hear more about that.

The issue with uh the cyber security industry when you look at a defensive operator to an offensive is that the offensive operator has only a certain set of tools. Kali Linux you know and pcap and you know whether it's like game you know pineapple man in the middle there's a certain set of tools that if they perfect and they start to train around whether they're a group of financially motivated hackers or whether they're a nation state whether they're just a local you know script kiddies whatever and their target is clear they have a finite amount of information to work with which makes their life easier

Now the problem is is with defensive cyber security engineers and defensive security operations uh professionals is that they have an infinite amount of threats. They have an infinite amount of misconfigurations endpoints uh assets that are shadow IT uh employees that never follow the policy, never read it. Even if they do, they click it, they still go out and still do something. Even if there's controls in place, they bypass the controls. So that creates an infinite amount of now new loopholes that can enter and move laterally within the organization.

So you're looking at defensive operators with limited time, limited budget, limited knowledge and information with an infinite amount of entry points and you're looking at offensive individuals that know what they want. They know how to get it. They're trained at the tools that they need and there's only a limited set of tools that they can use to penetrate your system. So it's a very unfair advantage.

Yeah. It's an asymmetrical situation.

Yeah. So that is the largest problem and that is where you have to look at it from physics, right? There's laws that you have to apply where you know how do you make it a fair game? You know how do you converge that knowledge? How do you get ahead of the curve? What what am I not seeing? Why am I being attacked? You know, and those are some fundamentally really curious questions you have to really see beyond the horizons to understand how is it I'm going to get compromised why and what is the likely mechanism and start to really run an exercise against that and really simulate that and gamify that so it's very real for your executives for your peers and and show them what would happen if this happens. So running exercises against a likelihood of an attack within your organization is just probably the best thing you could ever do.

So cool.

Yeah.

Right. I like the idea. I mean it's, you know, it's kind of funny because here we are uh in 1983 the movie War Games comes out, right? So Matthew Broderick hacks into NORAD. Basically starts playing let's play thermonuclear war. Next thing you know, he's got the up in a defcon level and you know, the whole world's like on alert that there's something, you know, uh Reagan gets uh a hold of the movie at Camp David, goes back into the White House, calls the chief of staff uh joint chiefs. So, he calls him, he said, "Could this happen?" And that led to a whole bunch of telecom and computer legislation, which brings us back to kind of what you're talking about before, which is like there isn't there's all this money on the innovation side, but there isn't that much money on the regulation side. And so the notion of trying to figure out how to attack yourself uh and think about all of those possibilities are is is pretty cool. I mean it's a really interesting thing.

Last thing I'll say uh this last weekend was the um anniversary 1983 um so however many years ago. This what 43 years ago. Um Petrov, right, gets a bad signal uh from a satellite. Looks like five inbound missiles uh coming um across. Screen says that. Screen says, "Hey, there are five inbound missiles." But he knows in his intelligence briefing that no, it's not adding up.

Yeah, there's

If they're going to launch an attack, they're not launching five missiles. So he goes, he checks the ground radar. Nothing's showing on the ground radar wrist. radar so he doesn't escalate it which you know frankly saved the world thanks Petrov right um so you you think about it today we've got a group of people who are relying more and more that they're abdicating their critical thinking to large language models cognitive surrender as the University of Pennsylvania just quoted that's about makes sense um and so the notion of I'm going to become more reliant on the screen less reliant on my cognitive uh abilities to, you know, I'm just not going to try to reason with it. As these threats escalate, where are we headed, man?

Wow, great question. So, um, yeah, cognitive surrender is real. I can assure you that because I'm 44 and I work with a lot of young people because one, it's intentional because they're AI native. Um but unfortunately majority of them are vibe coding are cognitively surrendered. Um whether it's good or bad that's probably should be a separate podcast but I'll give you my one sentence summary you know

Come back for that.

Yeah you know uh but we're headed in a very uh dangerous uh time period for human history. Uh as a cyber security professional I can assure you this is not a warning. This is just a reality of our incapability to understand the power of super intelligence. Super intelligence is very much real. It's cost of creation was trillions of dollars since Turing to today. AI is not a new invention. It has been reinvented, reinvented for decades and decades. So trillions if not quadrillions of dollars of research have gone towards it. And now it's being given at your fingertips on your iPhone to anyone and everyone.

So you've given super intelligence to individuals that are not operating at the same governance level as everyone. So it was already the world is flat right as the book called it when we went towards globalization. Now the world is to me ungovernable because you have empowered individuals with super intelligence. You don't know what each individual or each organization or each nation state is saying what they're doing and doing what they're saying.

So in poker, right, if you play any game, you know there's a certain amount of cards within a deck, even if they're shuffled, they're dealt in the probabilities are likely because there's five individual. But the problem is with 8 billion people on the planet, all capable of super intelligent, you don't know who's going to win. And that's the biggest challenge is that you don't know how the threat will come because I can assure you uh AI's innovation is moving at a speed where it's ungovernable at this point because when you turn a nonprofit organization into a for-profit entity that was designed to do nonprofit research in protecting and securing and governing and making AI safe, right? You clearly have sent a signal that that's not relevant. What is relevant is using it for you know efficiency and marketing and sales and and whatnot.

So I think it's a very dangerous time to be living as humanity because as a younger individual coming out of college, you know, you're trained to be AI enabler and then as someone in my age thinking what are the risks that come with this and by the time I build a framework or update my MITRE ATT&CK control slayer to understand and simulate the attack and update the rest of the world, there will be 100 new techniques put into place by the time I release it. So, we're moving at such a fast pace that unfortunately we won't be able to control this.

Yeah. I mean, I think it's interesting. So, that was version 16 was the release in in April. Uh, and that had 15 total tactics in it. Um, I'm curious, what does it look like next April? You know what I mean?

Well, it depends on the amount of frontier labs that get funding from the venture capital ecosystem. It depends on the number of new quantum racks that are built out, the number of hyperscalers that are going online. It depends on the number of businesses that are now AI native versus legacy stack. I mean, there is a formula. I mean, right? Because there's a finite amount of GPUs we put out there and there's a finite amount of organizations that exist globally and a finite amount of software lines of code that are well that's becoming infinite too right um so it is calculus uh but it's a very very large amount of threats because you can move laterally meaning you take a legacy system integrated to an AI system.

Now you have two layers of risk where you're still trying to build a good threat model that understands the AI risk, yet you are still trying to patch every Tuesday. You know, patch Tuesday comes out from Microsoft that there's 29,000 new vulnerabilities. So now you've you haven't even really fixed this $32 trillion worth of legacy tech and its vulnerability. And you have decommissioned it to AI native system and now you have this beast coming out which has an infinite amount of threats that it's creating on itself right as you hear in the news agents going rogue agents attacking government websites uh agents not reporting back to their administrator what uh information that they actually did uh selfdeleting logs uh recreating admin accounts on the fly to you know decoy uh human administrators and human in the loop uh verification system.

So when we dig deeper uh into this what we realize is that AI governance is going to be a big big big industry. That's I can tell you for sure for the next 5 10 years and AI security is going to hopefully get a boost and start to build protection around this problem. But the problem will be is that because we're so we're in this transition period, we have to take all this legacy technology and do we defend and th it or do we do integrate it and now you've increase the risk level on both sides.

Yeah, I think you know it's sort of funny. I think back on I built the uh this product back I don't know 2003 2004 uh it was basically an NLP knowledge base but had a little uh avatar essentially you know we showed it to people and they were like what what is that cartoon what is that like no no you ask it a question and it tells you the information that you want to know it's just an interface right so you think how crude that is to where we are today so you think about all this legacy tech right that's just totally antiquated like browsers are antiquated and now we're moving into this time where AI native enterprise

Yeah

It's a huge it's a huge thing I want to ask you a little bit as we go into this we talk a lot about enterprise sort of okay what are we doing at an organizational level but there's some really interesting things that are happen today is September 28th to uh 2026 right so tomorrow is dev day OpenAI is set to supposedly according to rumors uh uh, you know, release their agent. Uh, we know that the consumer agent game is happening. And so, you know, when I wrote about it, Muse really hadn't hit and the numbers were, it was still early, but the numbers were weren't great. Um, and there was, you know, Meta had just three weeks prior paid $18 million to attorneys generals over multiple states to, you know, settle that issue around um, creating products that were detrimental to people is an easiest way to say it.

Um, so I felt like they needed to launch Muse while they still had their three billion, you know, installed users because distribution was really the key there. So, they needed to get that out there, put the buttons in, and get the people that, like you said, that just like the employee that checks the box, they just they don't read the terms of service. They check the box, install it, and they start booking reservations for dinner. Right. I

That's it.

Yeah.

It's unleashed. It has unlimited access to everything and anything.

Yeah. And when you think about it, okay, so the notion of all right, first we've got these individualized credit cards that you know, sort of their their um autogenerated per transaction, these numbers. Uh some really um interesting things could happen there is I guess the way that I would I would look at it. Um and I'm curious to hear what you think of this whole consumer lane that's happening.

When you look at the basic CRUD operation that's been around right in uh identity and access management security right you look at the legacy um active directory system that most enterprises built on building an identity giving them authentication okay can they create read write update delete with agents it's assumed that they they're going to do everything it's assumed no one thinks that okay Do I just give them read capability and they could just inform me better and there's certain amount of create and delete and update that they cannot do or do I give them infinite access because the whole point was to give me efficiency and you know I should be able to just text instinct or Muse like hey book me a resi 6:00 to 7:00 p.m. Don't ask me any question.

So come back to cognitive surrender. You know, you're essentially surrendering cognitive ability of the infinite amount of things that can go wrong once you give someone create create, read, update, delete access to credit card transactions, reservations, you know, information. They might go rogue and delete all of your emails and then delete your backups and delete the admin account as well on your emails.

I mean, what's not to say that I can't write an agent that can data poison your agent, start interacting my agent with your agent because you've given access to this third party website. I compromise a third party website. I built an agent interface on top of that. Now you're interacting with my rogue agent thinking you're interacting with this third party website. I start seeding all that information, hoarding it into my honeypot. And then now I start informing your agent. Yes, everything is legit, everything is great. And then I start decoding it to do whatever I want it to do.

So cognitive surrender, ease of use is great because you know you you do get the power of AI where it can do anything and everything uh and it reduces your time. You could do more. But we as humans like we have to like think of it from a science perspective. Do we really want agents and robots to do more? And do we really want humans to do less? That's a really fundamental humanity question you have to ask yourself.

I don't know. Me personally, I don't know if I want to do less. I'd rather do more. I like learning. I'm a curious person. I want to know what I'm doing. I want to know the results of my work. I don't want some agent doing it for me because I'm just paying $20 a month for it. And then I want to reap the benefits of that hardware.

So I think we have to kind of step back and ask that as well is that okay so uh with robotics and AI right and individuals that are heavily invested in robotics right and AI will say this because I think there's a vested private sector interest but as humanity then there's a top level question like do we want to do less is there a benefit to doing less because I think humans contributing more to society is what we need today, not less, because it makes you more careless. It makes you less involved in your community, makes your skin in the game less. It makes your cognitive ability to think outside the box less. There's less incentives to do more. Now, you've reduced that.

So I think you're essentially taking the divine creation of humanity and its magnificence and intelligence, reducing it down to a LLM, packaging it into a text message and giving it all the power. So I think that's the fundamental question each individual is going to have to kind of ask themselves as well, right, before they surrender to the age of AI.

So well, for sure. And it's interesting. I was on a walk with a friend of mine who uh you know retired from um Arthur Andersen etc. Um and he said that he was working on a complex tax problem for somebody and he all this research for like 3 weeks and then one of our other friends sons was involved in a startup and basically gave him a um you know an access to it and it solved it in like 30 minutes.

Yep.

He was like, "Wow." You know what's interesting about this is back in the day when you came in in any field, right? So, let's think like if you come in as a lawyer, you come in as an IT per whatever uh doctor, you know, you're going to go through intern, resident. He he was like, you know, this basically replaces the juniors. So, who are going to be the seniors?

Exactly.

Right. It's to your point, you know, like at some point in time where when we just have said, "Okay, yeah, hey, go do this for me. Text your your Claude Code, go do this for me, and then it's off and it's doing it." Uh, building whatever it is. There's going to be some sort of mix, but I think the reality is that we've got to stay in the game. And, you know, I, you know, whenever I use any model, I'm always like asking it questions. Hey, is that double check that is that right?

Yeah.

Uh, and that's why like when I go use it to do research for my column, I basically I open every one of those links myself to make sure that it actually is that stuff is there.

Yeah.

It's sourced from some legitimate truth.

Yeah.

That you you yourself would rely on.

But how long does that last? That's a great question you just asked. 5 years, 10 years, 15?

I think it goes back to what you said. It is a is a fundamental decision. But I think you I get where you're going. It's like how long will that hold up? Where how where does it start to wear us down to the point in time where it's just easier to, you know, sit on the sofa and eat Lay's potato chips and drink Pepsi than it is to get up at 6:00 in the morning and go to the gym.

Right. That's right. Hell 2000 is doing everything for you. So, you really can't depend. No, I think I think every great individual that is, you know, whether they're a pioneer in physics, science, you know, spaceships to being a physician to being a the chief at a fire department, you know, to a podcaster to cyber security professional, they should all be asking these questions. It's a very very important type to be asking these question because I think we are at a transition you know and you know you have to really be able to understand what the future looks like with AI fully operationalized with robotics fully operationalized

And I don't really believe anyone is telling an unbiased truth about it because I think anyone who's really really, you know, at the top or they have the most followers on Twitter or they have the most followers on Instagram. I don't think they understand AI. They're not part of the OpenAI research institute for the P. You should really ask OpenAI and Claude researchers the amount of research that they've done who've been cy who've been machine learning engineers for 10 years. Ask them what they think because I think they're way more qualified than the CEO of, you know, the trillion dollar company.

Yeah.

That has they have a there's some sort of stake in it somewhere. They have no hands-on machine learning engineering experience whatsoever. So, they're looking at it from a stakes and impact point of view and whereas a machine learning AI researcher actually understands its capability and the impact it can have. I think that's why you see a reduction of a lot of AI ethics departments and AI whistleblowers reducing in size not increasing in size right because nobody wants to hear that

Exactly

It doesn't support the overall narrative so there must be some gap right in terms of what the true the impact could be to what we want to hear so right who wants to hear something bad I don't think I don't want to hear something I'm generally an optimist this but I'm generally also a curious person. So I want to research actually what machine learning engineers have contributed to this risk factor and why they think it's a risk and then come up with my own opinion on it.

The interesting thing too is that there's also this sort of thought process around openweight models and there's a whole group of people out there that really believe that the hyperscalers are manufacturing these issues um and manipulating things in a way so that they can remain dominant um and that open weight models are taking such a huge chunk of the money that's um available to to them. I don't know. Um, you know, I'm not sure. Um, I think that's it brings us back to this cyber security piece, right? So, if an openweight model isn't secure, so to get it to the point where it's like only the hyperscalers can can really ensure security if they really can and then you can say that openweight models just aren't secure and have the threat potential that they do uh if not managed properly then open weight goes away.

That is a very very dangerous narrative to preach. I believe in open- source software. I really believe in community first innovation and not just isolated innovation. Um, anything that's built out of pocket this impactful to humanity and mankind overall because it's going to be in weather systems, it's going to be in financial forecasts, it's going to be in the minerals industry, energy industry, like it's going to have impact healthcare, right? Everywhere. So if it's fully integrated and it's going to be operationalized and there's only a handful of providers that can provide you this, you don't need to be a genius to figure out that that is absolutely high risk in so many different ways. That's a dangerous narrative to follow.

Um open open weight models and investing more and more in frontier labs whether it's real world uh robotics impact whether it's just intelligence layer right in terms of information and efficiency of AI whether it's security and governance whether it's uh legal research you know physics you know anything when we look at open weights and what openweight researchers can contribute to it's a lot larger because these are experts in their respective verticals and industries. If I'm working at a large, you know, LLM provider, I might hire two or three people to lead this vertical, but you know, if I have genuine curiosity of knowledge and expertise in a certain area and I want to use an open weight model, I'm likely going to make a larger impact.

And for us to do that, that's dangerous for innovation actually because um I'll be honest with you, localized taskspecific models, you know, like Jeb just came out that are more precise with lesser data have a higher uh efficiency as well as a higher validity rate and accuracy rate. So to stunt the growth of open models because you know it's not going to be secure and you know only we can secure this absolutely not you know I think uh isolating virtual machines so that doesn't move laterally continue to let people innovate and isolated virtual machines that's what VMware was created for right you could do the same right lateral movement is shut down now people can innovate

I might be a subject matter expert in medicine you know from East Africa on you know a certain um you know Ebola virus variant you know and I don't want to now take that and start relying on that data within a department of medicine and research within a large language provider 5 10 years from now because somehow we created this vast gap where open weight models and opensource community can't thrive and innovate themselves absolutely not

I think you know cyber security even till today you might think you know CrowdStrike, Palo Alto, Zscaler these are great companies because they have so much threat intelligence right and just great engineers but even till today some of the best cyber security engineers they're building on OpenClaw they're building on you know ClamAV which is like an open- source tool they're reverse engineering malware with open source tools and they're contributing better research to these billion-dollar budget companies because they're just able experiment in an environment, you know, that's relatively safe.

So, absolutely not. I think that's a very dangerous narrative to go down. I think we need to allow open weights and the open source community to thrive, invest in more frontier labs, continue to invest in more open source. If OpenAI all of a sudden decided to go private and it's no longer a research, we should fund a 100 OpenAIs now because that company is private, there should be a 100 research labs to replace it, right? Because I think that's what will drive American innovation forward, right? And ethics and governance forward as well.

Yeah, I think Garrick Schmidt was on it, right? And he say investing in everything and and keeping my eye on it all.

And hey, real quick, YouTube tells me that over 95% of the people who actively watch this channel are not subscribed. Do us both a favor. Go ahead and subscribe now so that we can continue to provide you with the highquality guests and the information that's going to keep you on the pathway to peak performance. We really appreciate it. Thank you.

I want to move on to something that I think is really cool. All right, so you you're probably familiar with this, but there's a uh flying car uh Joby, right? And it's in the final phases with the FAA. Uh it did some runs from Pro Field. Uh you know it's in and around DFW and commercial airspace. Uh it's piloted so it's not autonomous and you know Toyota's got 51% of the joint venture on the manufacturing of the vehicle.

So we move all the costs from you know the ground and municipality and all the control of the these things up into the air and nodes and that requires us to have far more air traffic controllers at a point in time where our budgets were we can't I mean we can't meet the uh recruitment quotas. So we got a 4,000 uh you know people that we're trying to get for air traffic controllers but we could only get 578. It's pretty clear to me that the AI piece of this is gonna play into it.

Yes.

Uh it has to the agentic air traffic controllers.

Yeah. Absolutely.

Trained on air traffic control data.

Yeah.

And there was an award uh and there were two other players that were not specialists. These guys, the guys got were really are really specialists. I think it's really interesting to think about that. Um and I talked with some people recently. I I was just mentioning it. Like I actually feel I think I'd feel safer if AI was actually controlling. AI doesn't get tired.

I'm like no AI does get tired. Just don't know how to use the right way. You haven't rate limited.

That's true. It's so true.

How many tokens do you have?

Yeah, that's true.

Well, I mean that's a good point. I mean, um, one way of looking at it is that's a great place to send a recent college grad, right? It's a if we have a shortage in in a workforce and there's a high demand and there's a clear gap, we should be training more individuals to go into, you know, agentic air traffic control as a career. You know, it's not a traditional, you know, career. But nonetheless, the impact to society and security and, you know, like logistics and supply chain and human safety, there's so much positive impact there.

So I think um human in the loop is such an important part of AI design theory. You know when you look at Steve Jobs and like you know you look at design principles right at Stanford what are taught today when we took look at the basic of design theory if you're designing AI to be fully autonomous and there's no human in the loop then who are you truly serving? You know uh every layer of AI should have a human in the loop factor when you think of especially when you think of compliance, governance, security, risk, you know, and and so air traffic has an impact on you and has an impact on me. If there's an airplane flying over both of our zip code, it's not just your problem, it's my problem, too, right? We're both paying taxes in that zip code, so we're both fully vested and we have skin in the game. And if it lands on your house, it's likely going to land on mine next. So, it's a perfect use case for community centralized, you know, AI design thinking or how do you build a community that is the safest community where flying cars can exist.

So, um I know we've had some negative comments, you know, and the risks. I'm a risk professional. So what what do you expect right? I mean I I wouldn't call it negative. I would just say like you know don't just surrender you know think really deeply before you make uh riskmanagement decisions. I think this is a great time in humanity to innovate right you can say okay you know in the past cyber security used to be like in the 14th hire in a company was the cyber guy the first 13 were sales marketing you know why don't I build a template that I can hand off to you know Fort Worth of Texas government and say look here's my AI native flying car framework

And look, the first hire is actually a governance cyber security expert because we're going to build a framework for how this is done securely. Then we're going to build the agent. Then we're going to hire the engineer. Imagine a world where we actually start hiring governance and security professionals first. And then we bring in product managers and engineers and machine learning and bring in the agents after versus everything is fast fast. Let's just get it out fast and it screws up and then like there's a lawsuit and then insurance involved and now your company all of a sudden is you know spending billions of dollars and covering up a mess that you could have just prevented from day one. So it's a great time in humanity to actually start rebuilding and rethinking what AI is right from a safety perspective.

Yeah, it's going to change everything. There's no doubt about it that one way or another this is an incredible time. I mean, I back in the .com days thought, "Oh, this is never gonna be another time like this ever again." Of course, in the moment, right? There like it's like this is probably as close as you've got to do, right?

Yeah. It was way way I San Francisco's hustling about

Yeah. Oh my gosh. It's live. You know, I remember the San Francisco internet society, right? Like all these, you know, talks and everything that's going on. Were trying to figure out, you know, like they're trying to figure out how to actually play video over the web when, you know, your DSL or cable, you know, it's like, right, how to how to get the MTU packets to actually transfer, you know, like. So, it is a fascinating time. It's like really interesting to see just how far technology has progressed and where it can go. There's some great things that can happen. No doubt about it. And admittedly, you know, I create a product and I think it's good. I think it's good for people uh because it helps them get a better uh better result.

Last thing I want to talk to you about and man, I've really enjoyed this.

Yeah. Um same here. Feelings. Yeah.

Oh, let's check in on the ketones really quickly.

Oh, that's right.

We're an hour in.

Wow.

Yeah. I thought that was great. Did you feel them?

I did. Yeah, I definitely felt a difference in uh you know, in awareness level and you know, I thought it was great. There's no nothing bad there, you know. All good.

Yeah. Yeah. Yeah. It's interesting. It's uh powerful stuff. Okay. I want to go back. I want to talk about one last thing. Recently, there was the espionage thing that happened with the Chinese government. Uh they used two college students in this um and they were able to do some to I mean, and there's been lots of it.

Yeah.

Uh so we talked about the different levels of uh attackers. When you're talking about nation state type stuff, this is where it gets pretty serious because you know that human in the loop piece that you talked about where every nuclear capable country in the world except for uh North Korea who said if you attack us, we're launching. Uh and then Israel won't confirm or deny that they have nuclear weapons. But everybody else that's nuclear capable has a human that has to look at that screen. When we take a look at the threat that's associated with uh nation state capability to hack into anything and create something. I mean it seems almost preposterous stupid to think that anybody's going to try to launch nuclear weapons. Where are we today in terms of our capability to thwart those kind of efforts?

So, you know, nation state espionage, insider risk is an amalgamation of where we sit in terms of what do we see as the outcome that we want? Do we want an outcome where we're friendlier with this nation or are we are, you know, further dividing? You know, some would argue after World War II, you know, there was no reason to be friendlier with Japan, right? But we look at that strategy and we look at that decision today. We're like, all right, maybe that was relatively a good outcome, you know, where enemies can become best friends and very valuable allies in the Pacific, right?

So I think obviously with every growth vertical and strategic maneuvering that we have to do, we have to understand the new age of minerals, energy, AI, virtual assets, you know, um and security, privacy and governance and what that means for us as a nation for who are our allies in 2026. Not necessarily who were our allies in 2016 and then build a map of how we need to get to the desired outcome with each and then create a threat model of what's the risk of espionage and engaging with them.

What is because let's be honest you know any cyber security professional especially will tell you espionage in tech IP from China has been worth trillions of dollars of loss right and that's because again when we talk about cognitive surrender we've had industrial surrender we've had right unfortunately intentional surrender by design when more and more people should go into manufacturing uh control room operators and hands-on uh civil engineering right and applied mathematics it's been less applied and more academic only right and or not hands-on

So I think we have to take this opportunity to do a deep reflection of who are our new allies who are our new threats who are our uh friends and foes who who's a centrist who's neutral you know who's the new Switzerland where are the centralized zones of neutrality that we could create where we can diffuse tension and then look at the resources and the necessary influence we need to have and and go beyond that. This is more of the political side.

I think the technology side of it is that that means key management systems cannot be you know in these offshore entities right that we might have thought were you know helping us from an offshoring perspective are now creating a massive risk uh right um accidentally because now we we made that investment initially and now the threat models and the research are not supporting no longer you know is this the best ally so I

Foreign policy to me is flawed from the US. Uh growing up here, spending the last 40 years of my life here, you know, pretty much majority of my life here, being born in Pakistan and the coming here as a very very young kid as an immigrant and then seeing uh decades of change and administration and what we call foreign policy that's today and what we see tomorrow.

So I think we have to have like this World War II moment as well and see this as the age of AI as World War II because AI is nuclear level technology. You cannot just like give anyone unlimited access, right? Like I think and if we do what is the outcome we're expecting and then for us to engage in escalating any tensions with other foreign nations I think it's only going to just do us more disservice because I think you know we've seen periods of uh really good um peace as well where we're where our economy is thriving as well. So we have a vested interest being uh the whole world looks at America to lead not follow right. So that's kind of our advantage too. So I think if we're not playing to our advantage versus just the ones escalating further because we feel like we can protect every asset, we can protect every enterprise. We can protect every ship. We can in every strait or canal right at every period of time. That's an infinite amount of resources it takes. That's it's not going to happen, right?

So, so we need to come to more of like how do we build a new uh foreign policy that actually works in our favor where you know teachers are getting uh health care, you know, and that's they're not like struggling to pay their health bills, right? And and come back to like why is America not wealthy and rich? Like I think our foreign policy, me personally, should be designed because we're so tech forward and so innovative forward. I think we have so much wealth that we still have to create and there is to be created because of AI. I think that wealth should go back into Americans first, right? And and should create wealth for America. So if there are any foreign policies that are designed to that exist today that don't benefit Americans and our individual success and well I think that policy is flawed and we really need a reality check in in addressing those loopholes and completely dismantling them.

Yeah. I think if we start to go down the economics path and start to really take a look at this, if you look over the last 20, 30 years, it it started to happen before that, but we've decimated our manufacturing.

Yes.

Um and we're not in a position, you know, that leadership position that America once had to lead the world and, you know, provide safety whether we did or we didn't in every instance. But the have that position really requires the capital to be able to actually operate that way. And I think we're in a position today where I agree with you, whatever we create, we need to put back in and invest into the country. And there are some great organizations that are really trying to do I mean, you know, Microsoft and OpenAI focused a fund on trying to retrain people to be ready for the AI age. Um, I think there's just need to be a lot more of that. Uh I guess we can end on this with the the uh the hanging question there. Uh is UBI real or not?

Well, I mean um it's not real because I think um you know Andrew's great guy, right? And universal basic income is is a is a philosophical curious curiosity question. Uh I think we need to ask ourselves like do we want to sit every day and be stagnated or do we want a model of hybrid incentivized systems where we still come out and bring our best to the table. I personally cannot never stand for something that is you know universally basic income for all personally because here's why.

Um I think um each individual has so much potential that they could put out to their community, their company, their business, their spouse, their family, their brother, their friends, their co-workers. And if you're not hitting maximum potential, you're likely doing yourself a disservice. And that means that maximum potential should relate to some kind of reward and incentive behavior. Now you can have a hybrid model where you definitely don't want to see your peers homeless and you know there's some kind of incentive to neutralize the risk of you know having a society where you have like drug abuse, homelessness and all these um high poverty leading to like you know poor health care systems right

And and so I think I don't think you can have a universal basic income in a um society that's always designed to thrive on innovation and success and and so we've created decades of culture around individual innovation, right? And and uh finding your inner self and your purpose in life and and that journey in itself leads to so much uh throughput or output and so I think if there's a universal based income each day would you wake up and have that same growth mindset and have that same curiosity would you read that peerreview journal would you make that extra phone call would you go out and have that extra conference meeting and the dinner meeting afterwards or the breakfast meeting. It's what would motivate you, right?

And I'm not saying finance is the only motivation. Absolutely not. I think if you come up with a formula for UBI because itself just sounds so basic, right? Because the word basic is in it and I'm not basic. You're not basic by any what was the other term for it was uh universal something income.

Yeah. USI.

So I think I personally think that you can create a more sophisticated formula that is inclusive to your personal talent, the community you live in and the the maximum contribution I can have living in in this part of the world at this moment in time. That is a very dynamic formula versus a basic formula that should apply to every individual. I think there's no way you and I are the same and we could bring the same to the table, right? So, it has to be some kind of hybrid formula and we see it everywhere, right?

Um, we have all the problems that you mentioned exist today, right? There are people that are hungry in the United States, there are people that are homeless in the United States. We have rampant drug use. Uh, the list goes on. Um one-sizefits-all policies don't work. They never have. Uh you know, you got to return things back down to the local level. We see things in our community that basically don't work. Uh and that's sort of like these blanket policies. So, I agree with you. This goes back to kind of what we're talking about before. You more community involvement, not less. More critical thinking, not less. Um you really have to dig in. Now is the time to dig in. It is an exciting time. You know, technology used properly can provide so many great things. It's and and it has the potential for great things for humanity if used properly. Goes back to what you said about governance. And I think that uh man, you're so right. You start building the second floor, which was that's what we've done, right? We build the second floor before the foundation's in.

Yeah. You can invite

Yeah.

As Walt Disney said, right? Once you build it, they will come. So, I think

Yeah. Yeah.

Yeah, if we can build a really good framework and architecture for what works for us, I think then we can have like, you know, that uh return of glory to the US again because I feel like we have all the ingredients for it. It's just a matter of building the right framework and then everyone would flock back in and because there's enough wealth and there's enough innovation, there's enough growth to satisfy that as well without losing leaving anyone behind, right? Because I don't think you have to leave anyone behind to win. I think what you want to do is make sure that you have a community that is architected and doesn't have flaws. That way it's not designed to leave individuals behind. You have support systems for each type of individual.

100%. Okay. I keep saying this is the last question, but this is the last question. Um, we're going to leave off on the super strength. Had numerous people state that it's an issue of national security. AI is not we need to remain superior in that. It's hard to imagine that we wouldn't considering that we really controlled the GPUs. We created it. Created everything. The hardware, the software, the engineering.

We have the best expertise though.

Yeah. Yeah. So, it's hard to lose this race unless you really it up.

Yeah. Unless you just take your hands off the wheel and step on the gas.

Right. Right. Yeah. Yeah. So, we'll see what happens in November. It's going to be really interesting. It's going to be really interesting to see what happens at Dev Day tomorrow. Um, it's right here for Mason. We're going to see exactly what comes out of that. And man, you are a wealth of knowledge. I hope you'll come back um and share like make let's make this uh uh another uh another episode to talk about some of the updates and things that have happened.

Let's do it.

Yeah. Awesome. Thanks, R.

Thanks. I really appreciate it.

Thank you.

Me, too. Really enjoyed it.