Knack 4 Business

Human Habits Break AI Security: 5 Gaps Owners

Episode Summary

Michael Gershfeld and Michael Silva of Emely AI explain where your data goes when your team uses AI, and the five security gaps most owners miss. For SMB owners and leaders who want AI without a leak.

Episode Notes

GROWTH PILLAR: Cybersecurity & IT
WHO THIS IS FOR: SMB owners / Solopreneurs / Leaders building systems
WHAT THEY'LL GAIN: Five security gaps to check this week, a plain-English view of private AI, and real court cases that show why public AI chats are not private.

Most owners think AI security is a tech problem. It isn't. Until your habits change, no tool will keep your client data safe.

Michael Gershfeld and Michael Silva of Emely AI explain data privacy in plain words. Gershfeld brings 30 years in tech sales. Silva has 25 years in IT and cybersecurity and runs ArchonOne. They show how a private AI setup keeps company data inside its own walls. They share real court cases on AI and data privacy. They also point out weak spots most owners miss, like shadow IT, weak multi-factor setups, and forgotten service accounts. Their cyber awareness message is simple. Tools help, but human habits decide whether data stays safe.

Key topics covered

Guest links

Read the full write-up.

Supporting persons mentioned: Lisa Snow, who made the introduction.

Browse all episodes.

— PARTNERS ON THIS EPISODE —

Next Steps

Episode Transcription

Bernie (00:23)
Question of the audience.

what actually happens to your business data when your team starts using AI tools? And does your AI vendor even know? Most businesses, owners are asking the wrong question. They're asking whether AI can help them. The right question is, where does your data go when you hit send? Today, two experts who build AI from the inside out are going to answer that for you. And trust me, you'll want to hear this.

So we have two guests with us. And first up is Michael Gershfeld, also known as Gersh. Brings 30 years of experience in tech sales and business development, working with companies like Canon, Datto, Indigo, and before landing at Emely AI, and Emely is spelt E-M-E-L-Y, as the EVP of business development. Gersh is the guy who makes sure the product actually gets out in the world and into the right hands. At Emely AI, sales, marketing, and customer.

Gersh is just the guy who makes sure the product actually gets out into the world and into the right hands. At Emely AI, sales marketing and customer success all report to him. When he's not building partnerships, you might find him serving his community as a gabai at his local Shabbat or training Kenpo Karate.

Find him at Emely.ai. Again, I'll repeat the spelling is E-M-E-L-Y.ai. Also joining us today is Michael Silva, CEO, CISO, and founder of both Emely AI and ArchonOne, a legal focused cybersecurity firm he started back in 2010. Michael has a master's degree in cybersecurity from Northeastern University, Summa Cum Laude and he holds a CISSP certification.

He's taught over 1,100 students across more than 50 university courses at NYIT. He's an Eagle Scout, a third Dan Black Belt in Taekwondo, and a TEDx speaker. When he's not protecting businesses from cyber threats, he's at home in Long Island with his five boys and Rusty, the 100-pound Goldendoodle. Find them both at Emely.ai. Not the Goldendoodle.

Gersh and Michael, by the way. And gentlemen, absolute welcome. And you know, a pleasure to meeting the both of you. And do you have a favourite quote that you wanna let everybody know about?

Gersh (02:30)
Yeah, Michael.

Michael (02:31)
Sure. do it. There's less thinking, just do it.

Bernie (02:35)
Let's do it.

Michael (02:35)
Just do it.

Bernie (02:36)
So question to Michael. What was the moment that made you say AI has to have has the s security around AI has to be built differently? And you know what? that's a really true point because it's it's kind of a new area.

Michael (02:51)
Right. It's a really interesting question. So the answer is it has nothing to do with artificial intelligence. A lot of people look at the current state of AI and they try to retrofit things onto it. Human behavior hasn't changed in generations. Right. We still have thieves, we have liars, we have people who are deceptive. The tools are different now. And that's what we're seeing is that actual implementation of that bad behavior, and we're calling it artificial intelligence.

But the fundamental underlying driving attitude and the lack of integrity that some people operate with in this world is forcing us to embolden the or enhance, excuse me, the artificial intelligence secu cybersecurity safeguards we have today. but it's it's not that this needs to change or be added. We're just doing not a good enough job of keeping up with modern defenses against what is the most ancient h human bad behavior.

Bernie (03:39)
So is that the product development or the use of that you're guarding against or both?

Michael (03:45)
So it's a little bit of both. So remember the way, remember the easiest way to understand what modern AI systems are today is software. Software is automated labor. Labor should be a process, right, which you run inside of a company. And the process is simply manual steps. So what you're really focusing on is the ultimate implementation of the predictive automation behind what labor is. Right. So the question is really not.

Are you securing AI today? Right. It's more of like what is what is your process leading you up to it? Because one of the I think what a lot of people confuse is that leading up to that point where you're making the decision on what the AI system is producing and not producing is a is a probabilistic decision. Right? People like to make make jokes out of the outputs in these things, and there's disclaimers everywhere saying that, you know, this might be wrong and it often is, right? I posted about that on LinkedIn today.

But yeah, that's the answer.

Bernie (04:37)
So this almost bugs another question. Do you have AI watching AI and the end user then as a guardian?

Gersh (04:43)
I think the human has to be watching the AI and the AI has to be built correctly from the ground up.

Bernie (04:48)
It makes sense.

Gersh, you you came from a tech sales background from large large corporate companies. What

did that teach you about about businesses and what they actually fear when they hear the word AI? what what is what is the angst they're they're getting from that?

Gersh (05:04)
Well, anytime there's something new, there's always a lot of concern and people get nervous. it happened when we started adding computers into companies and networks and databases. it happened when we started having email and we started sharing files on the internet, and you know, it it it's happening again. and there's a reason for it. You know, you you're worried about

Who's gonna see this? Who's gonna have access to it? What's it gonna do? Right now, there are concerns about lost jobs and and the impact it has on culture and other things. it's really it's really quite remarkable because up until now it's always taken a long time. Now it's taking very little time. So the velocity of change now is something else to be concerned about. You don't want to be left left behind.

But at the same time, you want to make sure that you're not putting yourself at risk.

Bernie (05:55)
Wouldn't companies do a sandbox environment and trial is something it's much like, if you've come up with a new product, right? And a new airplane, new ship, you do sea trials before you go to we're we're ready to rock and roll full time. Is the sandbox environment not enough or is it is is the evolution of what's happening behind the scenes always like that that was a hidden surprise and we found a pothole in the road of life.

Gersh (06:18)
The the the the number of failed proof of concept deployments is still really, really high. MIT reported 95% failure rate a year ago. I think it's it's now like 87%. It's getting better, but it's still horri horrendous. So yeah, I mean they sandbox these things and very often they'll pull the plug because

Either it's not doing what they said it wanted to do, or it was a great idea, and it's doing what they said it wanted to do, but it can't scale, or it could scale, but it wasn't built with security. So there are a lot of reasons why things fail. for us, the the the real interesting question right now seems to be the what? What is it that we should be doing? So two and a half years ago, when we first started doing this together, it was all about execution.

Right. We just had great execution and that was a big differentiator. we can get it done. We could build this thing securely. Now the question is, what should you do? You could do anything. What should you do? What's the right thing to do? So that's really the beginning point. And when you talked about my background, the one thing that hasn't changed with all the technologies and everything I've been doing is the is the same thing exists where you have to find out where it hurts.

And what's worth working on. And then bring them a good solution and you have a business. You know, I mean that's that's what we do.

Bernie (07:34)
That brings me to the next point. how did you two lads join up forces? what what was the and yes, let's hold hands together down this journey of this yellow gr yellow brick road.

Michael (07:44)
Bernie, I I Gersh spoke earlier, he's gotta tell this story.

Gersh (07:46)
Let me tell let me tell you everything, and then you maybe you could talk about how we work together. So I was working for a company called Datto at the time that I met Michael. this was the preeminent greatest backup and disaster recovery solution available to anyone that had on-prem servers running. The solution was only sold through MSPs. There was no way to get the stuff direct.

And I was brought into the company to create a new channel, like a special projects thing and build out a whole new sales team and then another sales team and then another sales team. So one of my reps was assigned to Michael's MSP. and I just happened to be in the meeting for that initial appointment. And Michael and I just became fast friends in that meeting.

And then what wound up happening was the company wound up getting sold and taken private by another company called Kaseya. I gave them a year that turned out to be very painful. and at the end of that year, I decided, okay, it's definitely time to do something else. So I called Michael and I'm like, what are doing? And his response was, I can't believe you just called me. I was just gonna give you a call because I built this thing and I wanted to show it to you.

So Michael, why don't you take it from there?

Michael (08:56)
Yeah, sure thing. Well, and first of it's probably one of the best phone calls I ever made in my life. I maybe next to finding my wife. But it's been an absolute pleasure to to work with Gersh on on this here. So what's kind of interesting is what what Gersh mentioned before about our backgrounds. So

Gersh's background is as he stated, and he's quite good at that area of business, by the way. My background is in cybersecurity and information technology. So what I was building for was a system to make my job less of a pain in the ass. Right. So I've my background is in information technology, cyber. I've managed private IT departments, publicly traded companies, advertisement companies, IT departments, lots of different places. It's the same thing in every single shop.

Right, you you've got human beings coming in asking basic questions and you're the guy that happens to know the answer. When we saw GPT three five come out, we saw a compression of that knowledge exchange, but the human behavior didn't change. And people still kept coming asking questions. So the software that I was working on at the time was actually a prototype to help answer those questions because the the organization that I technically still run ArchonOne was an MSP, is a vCISO advisory firm for law firms in the city.

These law firms have banks as their clients. So what we found, what I personally had to deal with, was 300 question questionnaires from the bank's clients. And I would love to tell you that they were interesting, right? The only thing that was interesting about them is that they repeated the questions in the survey to see if you were bullshitting, to see if you didn't know what you were talking about or both. Or maybe you were lying. I don't know. But they would ask you the same question multiple times over to see if the answer aligned.

And each time it required a certain amount of diligence. Each time you couldn't be wrong. Each time that answer was specific to that client, to that particular law firm, right? And the answer had to be provided timely at scale. So the story goes that I I was working with the a managing partner at the time of a relatively decent sized firm that was growing. And he said to me, He said, You know, Michael, he said, you I'm very happy with your with your team's attitude. He goes, but your performance is just not where it needs to be. And I said

What do you mean? And I'm looking at the numbers and I'm seeing we're resolving help desk tickets within twenty minutes, we ninety-fifth percentiles. I pull the stats and I say, I said, Sorry, just to be clear, like we're meeting and actually exceeding our performance targets here. Like, could could you help me understand what you mean? And he says, Yeah, he goes, Well, this is just taking too long. I said, What what's too long? I guess we're doing this within twenty minutes, like ninety five percent of the time. And he says, It's gotta be fast. I said

Michael (11:14)
or what

Michael (11:15)
How much faster do you want it done? And he says, Three minutes.

And it was dead silence on the other because this is an attorney, right? So he's letting me figure the rest out. Just silence on the other end of the phone. And I said, Okay, let me break down this problem for you. Thirty seconds for the email to actually arrive in my mailbox. Thirty seconds for me to read it. Same with any of the guys on my team. Fifty percent of my shot clock is gone processing this message. And now you expect me in ninety seconds for ninety-five percent of our cases

To have that resolved within a certain CSAT. And he said, Yes, correct.

Michael (11:48)
I said That's what I didn't say yes

Michael (11:49)
I will work on this for you.

I didn't say no. And then that entire weekend I spent thinking, how can I get my guys to do this? And that's when it dawned on me, I said, there is no way a human being can do this. It has to be a machine. Put these

Michael (12:04)
Okay.

Michael (12:05)
two problems together and I started writing some software for that. What the the epiphany that that Gersh had

Michael (12:11)
which just.

Michael (12:12)
was the incredible catalyst to the whole conversation.

was when I showed what I had developed to Gersh, which was just a question-answering system, just for accurate answers, right? We needed to have grounded, actual, truthful answers. And Gersh looked at and he goes, Yeah, he goes, this is question for He goes, why why are you only using this for tech? why does this just have to be compliance? Why aren't you using this for HR and other organizations? I'm like, because I didn't really think about it, to be honest, that's the answer. And he said,

he

Michael (12:34)
You

Michael (12:35)
goes, there's a much bigger opportunity here for you. I'm listening.

That was the beginning of the conversation. So we launched this, and then what ended up happening after was GPT three five came out and we realized that we no longer needed to match fuzzy logic portions of questions. We could actually ask a natural language question. And that's when I learned about the incredible tsunami of BS that was coming out in this confidently incorrect answers from these language models. And I went, my god, this is incredible. I cannot believe how horrible these answers are. And everyone else was shocked.

That they got a better Ask Jeeves.com answer by staring at the mirror of their own data back in the modern world. And they were so blown away with what they said

Michael (13:12)
saw.

Michael (13:13)
that they bit the line and they were convinced. And we saw all sorts of horrible stuff come out of that. That's like any other invention, right? But we're not gonna blow up the plane as soon as we take off like the Wright brothers. We're a little bit later to the game. We're not doing Sora, we're not doing video, we're not doing, interactive 3D modeling heads, none of that stuff. What we are doing is secure, factual.

accurate grounded answers that meet these security and compliance targets, right? That's what we build for from the ground up. And this is where I was explaining to you before about the discipline that goes into building this. This is not something that you think about afterwards. If you architect your house and you build it out of balsa wood, you can't easily go back and wonder why somebody plowed through the wall like the Kool-Aid Man. This doesn't work that way. Right? You made the decision to not use reinforced steel and drywall or or concrete or whatever. Whatever you could afford

So this is the whole principle. And when Gersh talks about secure by design, secure by default, it is a mentality that has to be there. And the last thing I'll say on this is that it's not anybody's fault that they're designing insecure applications. it's your responsibility, but it's not necessarily your fault. There's the fourth quadrant of the unknown, unknown questions. When you're building anything, anything, I don't care if it's cybersecurity related or not, if you don't know the questions to ask, you won't ask them, you won't build, and the worst part.

Is these systems will confidently agree with you and allow you to trip over yourselves later. I and I could point to a million technical examples like this, but I don't want to bore I'm not sure if the audience is gonna be technical here, so I'm not gonna run them through the ringer of exact answers here. But just trust me when I tell you that it is far better to work with someone or ask the questions at the beginning. It's like pouring cement.

Michael (14:45)
Right?

Michael (14:46)
You get it right once, you never have to think about it again. Rarely. You screw

Michael (14:49)
If you

Michael (14:50)
it up.

Boy, is it gonna be a mess with the jackhammer masks later.

Bernie (14:52)
That legal counsel asked you, I need shorter turnaround times. Was he in the no or is he just pushing an envelope that he was trying to push and get a result that that wasn't already there? Because obviously if your competitors were doing it in three minutes flat, you would would have already known that. Right? So is is it is it so I'm going, let's let's let's squeeze a little bit here. Not so much f for the results, not not not so much for being

a knob about it, but you know, can we do better? Is is that always going to be impetus here?

Michael (15:21)
The answer is yes, but probably not for the reasons that you're thinking. So you gotta understand there's a certain so we're rather picky with who we work with.

Michael (15:27)
Right?

Michael (15:28)
The people that we work with are absolute animals in their fields. They are like the navy seals of the organizations that they are driving forward because they have the vision, they have the ability to execute, and they themselves have the personal leadership capability to drive the organization forward so they could pull like a vacuum even if they don't want to go.

Michael (15:44)
Right.

Michael (15:45)
What's happening is the leadership at the organizations we work with, and we won't work with guys that can't lead their teams, by the way.

Michael (15:50)
There's a lot of right?

Michael (15:52)
But the organizations we work with demand continuous improvement through progressive overload of their team's capabilities. It's like strength building, it's like any type of martial arts discipline, it's any type of hard work you gotta put in. It sucks doing it. That's why people don't do it. Because it sucks. You sit down and you think about it, and it's easier to just say tomorrow because that gives you the emotional payoff of not doing it today.

Michael (16:15)
Right?

Michael (16:16)
Certain

people who see what they want to get done and they say, We need to do this. And then what they they depend on their team that's working for them to actually do that thing. This isn't

Michael (16:27)
So this.

Michael (16:28)
this isn't this guy being a knob. It's not him saying, can we do better? Is it trolling me or whatever? I legitimately believe that this guy was a fantastic leader for his company. I legitimately believe

Michael (16:38)
that

Michael (16:38)
he wanted to challenge me to come up with it.

And I'm happy that he did. I do the same thing with with our team and we were when we're in a client engagement, we do the exact same thing. We ask uncomfortable questions. Everybody comes out stronger every single time. So I I think what he was doing was operating in his firm's best interests, which he should be. I just think it made everybody uncomfortable, which is fine. It's okay to be uncomfortable, Make progress.

Bernie (16:59)
Yeah, no, I I I agree. And just it was understanding where the person's coming from. Cause then if you have a client and they're going, Okay, this is the result, I need this to be better result. So always never be afraid to ask that question. Tweak it, can we? is the bottom line.

Now with AI floating around and people, I I know if I fire something into chat, GPT, Claude, et cetera, it you don't want to put your credit card information in there, your SIN number, you know, vital stats, you know, company secrets. But if you're using AI tools, where where is it set safe? Like it's is it just another piece of storage, like you know, a hard drive or you know, whatever wherever you're storing your data?

Except is it behind a firewall, is it on the outside of the firewall? Is it in a secure room? Is it air gapped security? Is it it's or is it just AI, just another another box of information that can be hacked?

Gersh (17:51)
It's it depends on the client's desire for architecture, but the way that we generally do this, the client's data resides in their environment and it interacts with a private instance of Emely AI in our Azure tenant, secured through their federated credentials and their security. And basically what happens is

Because we have a separate instance of the transformation tool, the rag pipeline tool, whatever it is that we're doing for the client with the AI on our side, it's a separate instance. So their algorithm, so it could be OpenAI or whatever, right? Whatever whatever we're using for them. It's a private instance of the algorithm. It's not like you're going to a public shared instance of the algorithm.

So because it's private, there's two things that happen. The first thing is we can we can guardrail it. And that's one of the ways that we avoid things like hallucinations. We can limit its knowledge to a private knowledge base. So when you're talking about things like policies, processes, procedures, resources that are required, things that you would have in in an employee handbook, an HR handbook.

those things. We have the ability to make sure that it's returning accurate responses because we're limiting what it's looking at. It's not looking at the whole trained model where there's a lot of bad information in there. The other thing that we have the ability to do because of the way we do this is we can control access to the information. So the thing that's really interesting about these natural language models, you're not.

accessing a document inside a folder. You're you're accessing a vectorized database that's generating a response in the moment. So it's a very different type of an experience. So if I've got this huge base of information that I'm drawing from, but I only want Bernie to be able to see this much and Michael to be able to see that much, and me to be able to see this much.

We have to deploy with role-based access control. And some of our models we're actually building with relationship-based access control. So the experience that the user has is always going to be based on who they are. And in some cases, who they know, what they're related to. Are they involved in this project? Do they know this person? those types of things.

So it really depends on the situation, but we have the ability to do very creative things without worrying about violating somebody's compliance requirements.

Bernie (20:24)
I guess that's the distinction between Emely and the say Microsoft Copilot.

Michael (20:29)
There's a little more than that. Let me just address the the privacy issue one more time. Just what's interesting.

Michael (20:33)
So.

Michael (20:34)
The the issue is actually bigger than I mean Gersh is is correct with what he said, but there's a bigger industry problem. So I'm not sure if you remember, but May thirteenth of last year, Judge Ona Wang issued a judgment against OpenAI with the New York Times lawsuit. Are you familiar with this lawsuit?

ChatGPT free, plus, pro, and Teams, just basically a whole customer base. The logged data that was produced and sent up to ChatGPT servers was required to be disclosed. So think about that. The only customers that weren't affected were customers that had a express zero data retention agreement. Again, go back to that unknown unknown question.

Hey, I'm on an AI system. it's not secure. it is. I went and I said d don't don't track training data. Don't no don't enable training on this. It's fine. It's not training on my data. I'm paying a lot of money for it. Have a teams agreement. It's thousands of dollars a month. We're not they're not training on our data.

Michael (21:28)
Okay, so.

Michael (21:29)
You

have to understand architecturally how the systems work, or you need to know somebody that understands because what happens is later you've committed all that data, later you've presented the information, you've stored it somewhere, and then through n no no fault or control of your own, your hand is forced.

Michael (21:45)
Right.

Michael (21:46)
So it's more than just this. Now you might answer like what's the answer to this? Right. I'm sure that some of our listeners are gonna be wondering, how do I deal with it? you just say professor silver.

Bernie (21:53)
Everything on paper. Everything on paper?

Michael (21:55)
you

Michael (21:56)
Everything on paper or you ready for it? Drum roll please. Encryption.

Gersh (22:00)
Don't retain the data. So like think about this. If I if I'm if I'm an ex employee who's suing my employer for discriminating against me, they fired me and now I I have a discrimination lawsuit. And my law firm now goes after them. And part of their discovery is I want to see his direct manager's chats with AI. I want to see what kinds of questions were being asked during the period of time leading up to this this employee's

Separation from the company. And now they go in and they see that there's like damning information in there. That's not good for the company. So in most cases, when we when we do this, we're not retaining anything except for the log. All you can see is that somebody asked a question, question was was you know responded to. That's it. You can see that data was moved, but you don't know what the question is. You don't know what the answer is.

Michael (22:50)
Yeah, there there's another component to this too. So Gersh's correct. That's the technical control. That

Michael (22:54)
Okay.

Michael (22:54)
is the other side of it too. Even even if this is where it gets really frustrating, right? Even if you're using systems like that that are super secure. There is currently, I'm not gonna say it's law, right? We talk about precedent. it was on, I believe, February 10th of this year, there was

Michael (23:11)
with judge.

Michael (23:11)
Jed Rakoff here, but we're in New York, right in the Southern District of New York.

ruled that documents that a criminal defendant generated using a non-enterprise consumer version of Anthropic's Claude were neither privileged nor protected as work product. It's the first, it's it's literally the first decision of its kind. So, in other words, like if you plain English, privilege requires a lawyer, but if you're giving this information to any other system, ChatGPT is not a lawyer, right? We've seen that in news. Okay. That's the end of the analysis.

So understanding how to work with this, right? So it's it's not here's the thing that I think should really terrify every lawyer, which is the the killer fact. Sharing privileged information with the third party AI platform may constitute a waiver of privilege over the original attorney client communications themselves. So let me state let me state that again, just so understand. It's not just that the AI chat itself isn't protected, which is important, like Gersh is saying, okay?

Putting clients' confidential data into ChatGPT blows up the privilege itself on the original client client lawyer conversation that it came from. So when confidential information from your attorney is shared with a third party platform, again, outside your hands, right? It can potentially waive privilege over the underlying attorney client privilege itself. So let that sink in for a second.

Bernie (24:25)
It's almost like you're crowdsourced and whatever's in the crowd and it's in the air, it's out in the wild for all to see.

Michael (24:31)
Right. So court courts have addressed the the privilege in similar cases, right? insecure email. I it was the United States versus Hamilton. The courts found that emails that were sent through an employer's monitored system were not privileged because they allowed inspection, which is the same logic as ChatGPT, right? You're sending data to a platform that openly collects, stores, and shares your data. So therefore, there's no expectation of confidentiality. It's a problem.

Bernie (24:54)
Well, so Emely just tracks that the date and time and an answer came back and then the person sees the answer. The person prints the answer inside their four walls then or captures it and pi cuts and pastes, which typically happens.

Gersh (25:08)
always save it someplace else. So if you want if you like that chat, you need it, you want to generate something from it, yeah, you can you can save it. but now it's it's saved on your computer. It's not saved in my ecosystem. Or you're saving it to someplace else on your network. It becomes just like any other you know point of data you know in your ecosystem.

Michael (25:29)
Yeah.

Michael (25:29)
But

Bernie, look, the the funny part is that paper is actually more discoverable than digital, not less. Right? Every every law firm in America has bankers' boxes from cases that closed 20 years ago that they can't throw out. So the answer is not going backwards, right? It's going forward with the right architecture. It's encryption at rest, it's encryption transit, zero data retention by default, a private instance where you never want data to leave. And and I think look, what you're saying here is, you know, with regarding the cut and pace point, I mean, this is this is correct, right? I mean, this is the human in loop problem, and this is the real frontier that we're dealing with.

We can lock down the technology completely and we do, right? But we can't lock down what someone does with the answer after they read it. And this this is why as a consultancy, we don't just sell software, right? We engage with the leaders to sell process. The technology may have 40% of what makes this work. We see statistics about success and failure and all that. But you can't patch human behavior. And this is why I'm explaining to you that we only work with leaders that will actually

make change in their organization because we cannot. We ourselves will fail if we go into an organization and say, you have a process problem. Nobody likes to hear that. Nobody. Everyone's seen office space. Okay. Nobody likes to see the process consultant come to you have a broken problem. you don't know my business. Okay, guy. Fine. Well guess what? Human beings have been around for a very long time and I recognize this behavior. So you have a process problem. Somebody at leadership has to enforce the new direction. When that succeeds,

Michael (26:51)
You

Michael (26:52)
it takes off.

When it fails, everybody looks around and goes, Hey, it's the new guy around here. He's the problem. Get him out of the room. I don't like the way this looks anymore. And what happens to the organization over time? It atrophies.

Michael (27:03)
Right?

Michael (27:03)
So this is where this is where this clear delineation needs to be made with artificial intelligence and the way it's actually being used. Right? It's not just the the paper element. It's not just the how do I keep this thing secure? It's the human. Everybody forgets about the human. Everyone's so focused on the tech.

Michael (27:20)
Right?

Michael (27:21)
It's actually about human behavior at the end of the day.

Bernie (27:23)
As a sidebar, because right we're talking about someone sitting at a terminal somewhere, or can you guess my age? Basically using the word terminal, but sitting at a keyboard or or sitting on their mobile device. Got that, got all that all that content. But in our lives, we've got more devices like right now. If I if I say hey and mention a name, the ears come on and it says, Yes, master, what would you like to know?

you know, or I have, you know, a Roomba, or, you know, and I'm I suspect from what we're seeing in the technology field that robotics androids are going to be moving around, or or even drones, you know, whatever the whatever the case is. Where does that kind of fit into the scheme of being secure? Because if you're doing a corporate, like I used to work in a healthcare facility, had 10,000 people in the box, another 10,000 people wrapped around it.

I watched the IT group go, yeah, you can do this, you can't do that. We do our upgrades slowly because, you know, everything you have to make sure everything fits still and operates. But as the technology wrapping around it becomes more sophisticated, RFID tags and then you know, something to collect the intelligence or CCTV footage. How does that all fit into that, into that keeping it tight and secure?

Michael (28:33)
Yeah, I love this question. Yeah. I'm sorry to interrupt guys.

Bernie (28:36)
No, no, it's all it's okay. It's just as you're talking, it's like hang on. next to me comes Chiefs, he goes, Hey, boss, what do you want to do next?

Michael (28:43)
Yeah, so I love that question and I w I can actually point back to where this has already happened. Right, which actually was January twenty seventeen in Dallas. Megan Neitzel's six year old daughter, Brooke, ordered a hundred and sixty dollar KidKraft Sparkle Mansion dollhouse and four pounds of sugar cookies by talking to Amazon Alexa saying, Can you play dollhouse with me and get me a dollhouse? So San Diego's I think it was CW

Michael (29:06)
I

Michael (29:07)
six covered it.

And the I think he said like, I love the little girl saying Alexa order me a dollhouse and guess what happened? On television when Jim Patton, who's the news anchor at the time,

Michael (29:17)
said, girl. What happened was

Michael (29:18)
I love the She said Alexa ordered me a dollhouse. What happened was

every single Alexa powered echo device in San Diego bought a dollhouse.

Michael (29:28)
Okay, so.

Michael (29:29)
And by the way, this wasn't the first time. In in in in 2014, Microsoft had an Xbox commercial with Aaron Paul demonstrating Kinect Voice. You guys remember Kinect? The Kinect Voice Control was blamed for causing consoles to spontaneously boot up and launch Titanfall. Can you believe this? So look, you don't have to go too far back into history to understand. Yeah.

Gersh (29:48)
Directly in Michael Michael's story is is is hysterical and and very prescient. You have this constant tug of war occurring between convenience and security. They're opposite ends of the same rope. And the Internet of Things is all about convenience. So, like all of the things you mentioned, like all of these automations that we have in our homes now, it it's all about convenience. People need to understand that.

You better be thinking about security. And I'll give you an example. There was there was a story about a a I don't I think it might have been in Las Vegas. There was a casino that had this huge fish tank. And the heater in the fish tank was one of these internet of things controlled through the network. And a hacker gained access to their network through the fish tank heater.

Because when they deployed the heater, they didn't change the passwords. They left them as a default password. So if you see that brand of the heater and you go to the manual online and see the default password and try the default password, you have access to the heater. And that's how they were able to backdoor in and hack into the network of this casino. And it happens on a government level.

With things like turbines and the all of these things have the same type of vulnerability if you don't change the password. Now it goes well beyond just changing the password, but that is the type of stuff that's happening because people are just not paying attention. So every sing every single thing that you put on your network, in in your in your business, in your home, minimally don't stay with the default password.

Michael (31:22)
Yeah.

Michael (31:22)
Don't

use a password at all.

Bernie (31:23)
Well, I I guess two questions come out of that. th that sounds like the low-hanging fruit. Like, you know, someone does, you know, almost like doing a drive-by going, you know, the old terms war driving, I believe. You know, you're picking up Wi-Fi passwords, right? To to breach things. What's the next level up that most people miss? And you know, whether it's I'm using AI or

And I guess these are things you probably would address. You do an analysis of, where where are your vulnerabilities? What what's the next level up of vulnerabilities that people tend to ignore or maybe paid attention to ten years ago and but haven't, you know, changed anything.

Michael (32:00)
That's a really good question.

Michael (32:01)
So.

Michael (32:01)
The answer is that ultimately this comes down to

Michael (32:05)
you

Michael (32:06)
the same it it's a mindset. It's it's it's a mindset ultimately, right? So you have to remember that the security controls that are put in place are only as effective as the person's gonna use them. And I I I I often demonstrate this with a very simple example. Everybody

Ye y the first of all, the reason why people react is security, typically if they're not if their hands are not being forced by compliance. You wanna know the number one reason why someone does something? If it's not being forced on them by by someone in compliance? What's the number one reason you think that people will actually do something with?

Michael (32:33)
like

Bernie (32:33)
They've been breached.

Michael (32:34)
Right, some they got their head blown off or they watched somebody get their head blown off, right? And all of a sudden they come out with this monster budget. And it lasts a very short time. It's like a firecracker. It's a beautiful thing to watch. And then suddenly you get all these rare tools in place. And what are you left with? The person that needs to actually implement it. So if you go back to thinking of the person who has when you leave your house in the morning, okay, you have a deadbolt and you have a key and things like that. Honestly, you don't have to answer this question, right? Because this is a recorded podcast. So how often do you really lock the deadbolt? Honestly.

Right, if it's not a hundred percent, now you are voluntarily accepting a lower standard of security. Why? Wrote it on

Michael (33:09)
Right?

Michael (33:10)
paper, you my god, I would never do that. I would never put my family at risk. All my belongings are in this house. I I I'm secure, right? No, not quite. You're actually doing what every human being normally does without thinking about it, is you're making it just hard enough to move on to the next guy where that attacker will leave you alone.

Michael (33:25)
That's what security is.

Michael (33:26)
Right? It's not meant to be completely impenetrable. Some things appear to be mathematically, the underpinnings of cryptocurrencies and other algorithms that depend on infinite compute time, but you've got quantum compute right around the corner to directly competing for that. So it really

Michael (33:40)
All right.

Michael (33:41)
comes down to the convenience. And and I think if I was gonna go back to the next level vulnerabilities, because I would love to talk about like war driving, by the way. There was a thing in Germany. I know you heard about Germany, right? Like what happened out in

People were driving around in I think it was

Michael (33:54)
Thanks.

Michael (33:55)
I think it was in twenty thirteen that Google got fined for this. Did you know about this? I'm gonna answer you in a second.

Bernie (34:01)
Well yeah, they no, they were going around picking up Wi Fi I

Michael (34:05)
.

Michael (34:05)
Yeah.

Bernie (34:06)
it had anything to do from the don't take their picture of their face.

Michael (34:09)
As it turns out, as it turns out, there are ways that you can identify where you are in the world with supporting data, some of the supporting data being the names of the radio signals that are around you.

Michael (34:21)
Okay, so.

Michael (34:22)
If, for example, in very busy areas of New York City, there is a fixed combination of network names that never really change that are only present in one geographical area.

Michael (34:31)
Okay.

Michael (34:32)
Therefore, if you're inside of a building,

And you pick up these Wi-Fi signals and you have no access to GPS, but you have a very specific list of network names that you're very close to, there is a degree of confidence that you can be sure you're at this particular

Michael (34:46)
Thank you.

Michael (34:47)
location with no satellite coverage.

Michael (34:49)
So.

Michael (34:50)
Wi-Fi goes everywhere. Okay, so you're driving down the street and you can pick up Wi-Fi. So companies that are in the data collection business and advertising business, Google, by the way, advertising company, so we don't forget about that.

In in twenty thirteen, Hamburg fined Google for this. It was, I think it was 145,000 euros for Street View data collection. If you've seen these things, it it's like these antennas sticking up because they're they're sweeping for this data. Now people love to play jokes about this that they put a bunch of cell phones on the road and move real slow and a little, Red Rider truck and simulate traffic conditions. you, you you can have a little fun like that. And I don't think you should do that for the record. But but the point is that when that data is being broadcast out and you're collecting it.

It raises a really, really important privacy question, which is if you're sending this information out, then do you have a are you do you have a right to privacy with that? Right? Like is it is it a reasonable expectation if you're broadcasting that out? it's a bit of an open question. I would, you know, human being to human being, if it's not your Wi-Fi network, probably you shouldn't be using it. But this was where security comes in, right? If you're broadcasting it out, a lot of people forget that the Wi-Fi connection is the exact same thing as a

Local connection plugged into your your switch. You wouldn't allow that. You wouldn't allow somebody to walk up to your house and just physically plug into your network, but you broadcast Wi-Fi out. In fact, you crank up the broadcast wattage so far because you want coverage on your front lawn that you, the next block gets it, which by the way is illegal for you to jack it up that high. But look, to answer your original question about the the biggest threats that I think most people miss, I would say I'm just gonna give you like my top, my top five.

Number one, identity sprawl. People have multiple accounts for things. So I'll give you an example. or just a hypothetical

Michael (36:21)
So, I...

Michael (36:22)
situation. If you have an employee at a company and they have a username and password, maybe a VPN, and somebody leaves the organization, but they're signed up with their personal email, the company may not have an inventory of that. So when somebody leaves the company, IT might disable their Active Directory account.

But that ex employee is still gonna have access to Figma, Notion, Asana, ChatGPT, Claude, Loom, you know, twelve different Slack workplaces, whatever, Dropbox folder. And we you know, I see this a lot in security audit work that we do. So that's a problem. So the identity management is a huge

Michael (36:54)
issue out.

Michael (36:56)
The second of five things is is Shadow IT explosion. This is a huge problem. Right. So in twenty fifteen, Shadow IT meant somebody bought their own Dropbox subscription. And, you know, in in twenty twenty six today, it means your associates are pasting client documents into fourteen different AI tools you've never heard of, which we just talked about, you know, with with no enterprise contract, no data protection agreement, no audit log. And it it's a problem. The official AI policy is a word document that nobody reads.

So this is the a real modern version of the problem that I saw back in twenty ten, except data exfiltration is voluntary and the employee thinks they're being productive.

the third thing would be multifactor authentication gap.

Michael (37:36)
Okay.

Michael (37:36)
This got paid attention to ten years ago and is ignored today. Most folks

Michael (37:42)
Thanks

Michael (37:42)
authenticate through either SMS or an authenticator app, or not at all, which is awful if you have no multi-factor. The problem is MFA has been considered broken since twenty seventeen. NIST is actually saying not to use it. What we really need to be doing

Is we really need to be using passwordless authentication, right? We need to be using device-based authentication. There's there's more here to this, but having simply a password and multi-factor, multi-factor is okay. And it's better than not having it. So don't read me the wrong on this. I'm not saying don't use multi-factor, but I'm saying if you're relying just on a password and multi-factor is a part of that equation, but you've got your SIM card being your two-factor, and somebody calls AT&T because I don't know.

Your social security number got breached and is now public, which by the way, it probably is based on recent breaches. And your name is out there and your address is out there. And all this other information's been leaked because you put it onto a bunch of different websites to sign up for services and their back-end data services got breached. And now people are stitching together a very clear view of who you actually are. And then they decide you're worth something, so take

Michael (38:46)
They did.

Michael (38:47)
over your account. And now they have your cell phone number. Okay.

And you're busy now trying to get a hold of the cell phone carrier saying, You guys have held my account hostage. How do I get back in? And the carrier is saying to you, I can't authenticate you, sir. And somebody else is receiving two factor codes to your phone.

Michael (39:01)
How do you deal with that?

Michael (39:02)
That's no longer a biometric check. So multifactor

Michael (39:04)
right?

Michael (39:05)
isn't necessarily bad, but it's not everything that it needs to be today.

Michael (39:10)
So.

Michael (39:10)
If you're not using multifactor at all, please turn it on. Like take this advice, turn it on. But it's not the end state. Right? More complicated answer. I would have to say the probably the third

Michael (39:20)
third.

Michael (39:21)
fourth thing, service accounts and API keys.

Michael (39:24)
huge.

Michael (39:24)
Most modern organizations have dozens of service accounts.

most ransomware attacks don't break in through a person, they come in through a forgotten service account with admin privileges that's not been changed since twenty nineteen.

Michael (39:34)
So...

Michael (39:35)
If

you're organization running this, you absolutely have to make sure that you understand what service accounts are running. there's ways to handle that, right? Privilege authentication. And I hate

Michael (39:43)
to say it this way.

Michael (39:44)
but one of my favorite new topics that I love to discuss, which I hate that I even have to have this conversation, is the supply chain blind spot.

Michael (39:51)
Okay, so.

Michael (39:52)
If you read carefully in the news, you will see that attackers are including in automatic inclusion of viruses into packages that are dependencies on other software platforms. And this is where I think probably it's in my top three of worst cybersecurity observations, like meaning, like I can't believe I'm seeing this, but I actually am. So when you're building software and you add content, okay, you're not gonna write all that code yourself as normal, right? That's why Microsoft

bought or not bought thirteen billion dollars nonprofit OpenAI, right? Why would they do that? Well they want them to play around with the GPT store and have them take the risk and then copy and paste their code because that's what Microsoft does, right? They're a licensing company. They want to develop everything. So that's fine. That's cool because Altman's going to treat them nicely and investigate the source code and all that. But hey you're not Satya Nadella and you don't have $13 billion. So what do you do? Well you go

Michael (40:38)
GitHub.

Michael (40:38)
right because look, there's that really cool software package that does just the thing that I need. So you download it and it's

Michael (40:44)
Safe.

Michael (40:45)
And you run the security sweeps on it and everything's good. And it it passes all the tests and everything stays fine for a very long time. And then you build and you build and you build new features. And then two months later, you look at the software and you get the security bulletin. And the security bulletin says, That package was hacked. And you're like, my God, I am so glad that I downloaded this version before it got hacked. Because I would have been absolutely screwed. And then you move on. And then a month later, you're breathing.

And you trace the breach back to that software package and you go, how'd this happen? And you find out that the person that was running that package, briefly,

Michael (41:19)
had.

Michael (41:19)
a virus included as a part of their build, and your software auto updated it because you were deploying new code at the time and you said, I want to keep all of my libraries up to date.

Michael (41:27)
Okay, that's

Michael (41:28)
Supply

chain attack today.

It's a problem. There's process. I

Michael (41:33)
Right?

Michael (41:33)
was telling you about this before. Software development security process you have to have in place to prevent this. If you're not working with somebody that understands these things and you're yahooing it, you are going to get screwed. There is no other outcome. Because the attackers are getting better at their attacks, it's less expensive to run. And quite frankly, I think it's more entertaining for them.

In some sick way.

Bernie (41:56)
Well it it might depend who the target is. that said though, So the Emely AI product then, at the end of the day, it looks at the information that a company wants to move back and forth. Does it also handle, you know, by the way you're at risk? Does it do a a validation? Does it do a cross check? Does it provide

Not replacing the the keen mind of the security expert or the architect, but does it support and go, something's not right here, boss, or y there's there's a wheel gonna fall off in in a in a few minutes from a tech perspective. Does it does it

Gersh (42:28)
It's

a more it's a more bespoke solution by client. So what we're doing, what what if you want to look at product, the the product is is our secure rag pipeline that we come to the table with if it's needed, when we're working with our clients. So it just gets us there faster and it gives them access to that secure environment where they can have that natural language experience.

Without worrying about security. The product actually is the consulting gig. So every single project that we've worked on is unique. All the clients are different, all the problems that we've solved for them are different. So it's it's really more a tool and an environment than it is a product. Now we are working on other things that will be product, but that's coming in probably another month and a half, two months.

Bernie (43:19)
Wow. Gentlemen, you know what? This this is a really I've enjoyed this conversation. because what about this and what about that? And here here we go. Michael's got, yeah, that that too matters.

You're actively looking for recruiting partner, channel partners, consultants, and MSPs. What's a good fit look like for you guys? And where?

Gersh (43:37)
Anyone

that's working with with clients who are producing at least ten million dollars a year in revenue, that's the one the one thing I could say, most of our clients are considerably larger than that, but the smallest are at least doing 10 million a year. they all have a a desire for security, privacy, and compliance. They have leadership that can manage change.

And they're looking to improve their processes. And and it could be the customer experience. It could be something internally. it could have to do with human resources, it could be recruiting, it could be retention, it could be they're they're looking to have an exit strategy ten years from now and they want to incr improve their valuation. It's it comes from all over the place.

the common thread is they have money that they can spend. They have real pain that is a good fit for us. And they're concerned about privacy and security and compliance for whatever reason. Could be they want to protect their intellectual property, it could be their clients are heavily compliant in financial space, healthcare space. It doesn't really matter what the motivation is as long as those things are in place.

We do work direct and we do work through channel partners. love it when when we develop a channel partner who really gets it. we value relationships and it's important to us because we don't do any outreach. We're one hundred percent inbound marketing, it's all referred lead prospecting, a hundred percent. And so we we compensate for for good relationships. Let's put it that way.

Bernie (45:09)
Are you continental US, North America, global?

Gersh (45:13)
Well, we're we're working out of New York. we're expanding certainly past our borders. we are we are capable of working anywhere where Microsoft has a data center.

Bernie (45:24)
So, folks, if you're listening to this podcast and you want to see Emely AI in action, book a demo directly with the team, head over to Emely. It's spelt e-l-y dotai. And hit the pick a time to meet Emely button and have that kind of take a peek under the hood. so if you from today's conversation, if you're thinking where your business data is actually going and you kind of need to know that it's tight and secure, in other words, you know whether you have a compliance issue.

That's mandated or know, you just don't want to get sued. Or it's mission critical. it could be infrastructure, you could be healthcare here in Canada. It's you know it's a public health care, but same thing. Reach out to Gersh and Michael's team. And I also want to say thank you to Lisa Snow for making this introduction. We appreciate the this cross connection. And just remember that Michael and Gersh solve a problem that most business owners don't even know they have. And in fact,

Michael probably can help you out on a lot of other areas, as in you better check that weak spot out. And your MFAs are are you know they're nice, but they're not the absolute lock on the door that you've ever been looking for. So keeping your AI safe is the key. And also keep in mind, if it's in the public sector, it's it's vulnerable, even though you might not have your information supposed to be confidential. So just keep that really strong point in mind.

to reach us is knack4business dot com. It's knack k n a c k the number 4 business dot com to reach us. Use our chat bot to find out about other episodes or book a call.