Bold Security Rethinks DLP With AI on the Endpoint
From the cloud back to the endpoint
Krasner started in cybersecurity at 18 in the IDF Intelligence Corps. He then moved through product roles in offensive security, threat intelligence and cloud data security. He and his co-founder, CEO Mati, previously built a DSPM company, which they later sold. While selling that cloud product, customers kept asking the same question: what did a user do with sensitive data after downloading it? As a cloud tool, they had no answer. That gap pushed them to found Bold in 2024. Today, more people work with sensitive files locally, run AI models on their own machines and feed more context to local agents. As a result, Krasner sees the device as the place where data risk now concentrates. Endpoint DLP is how he wants to manage it.How Bold builds endpoint DLP around AI
Because Bold started after AI was already in wide use, the team built it into the product from day one. Bold runs language models directly on each device. They handle data classification, lineage tracking, obfuscation detection and anomaly detection. It also recently added an intent agent. In effect, every endpoint gets its own agent that weighs context and makes decisions on its own. Security teams can still manage it centrally. However, there is no cloud round trip to fetch logs or ask a bigger model for a verdict. So users work without interruption unless there is a real breach, and the approach scales easily.Why endpoint DLP needs context
How does Bold tell normal work from insider threats or runaway agents? The answer, Krasner says, is context. Bold considers the user’s role, where data came from and who edited it. It also checks where the data is headed and even which account is logged in. For example, it can tell a company ChatGPT account from a personal one, and even tell subscription tiers apart. Krasner also stresses that companies must share sensitive data to do business. Many DLP tools treat every outbound file as an incident. Instead, Bold asks who is receiving the data, what the business purpose is and whether it fits past behavior, then steps in only when needed.Getting started with Bold
Bold offers a virtual environment for a first look at its endpoint DLP. Still, Krasner says a short proof of value in your own environment shows the most, since the agent quickly learns your data and risk tolerance. To learn more, visit bold.security.Transcript
Hey everyone. Welcome to another Techstrong TV interview. My guest for this interview is Hadar Krasner.
Hadar is the co-founder and CPO for Bold Security, and let's welcome Hadar to Techstrong TV. Hadar, how are you, man? It's great to have you on here.
I'm great. Thank you for having me. My pleasure.
So Hadar, I mentioned you're the co-founder and chief product officer over at Bold, but give us your path. How did you wind up here? Yeah.
So, I started my path at the age of 18, practiced cybersecurity in the Intelligence Corps in the Israeli IDF, and then from there, I segued between different roles in the product management position. So it started with a bit of cyber offensive, going through threat intelligence, and then data security in the cloud, and it all pretty much constantly pointed towards the endpoint, hence why we started Bold. Really.
Very cool. A lot of people think it's been about the endpoint for a long time. I myself have been in security probably more than 25 years, and there was a time where it was the endpoint, it was antivirus and host-based and all of these things.
But then, people started getting into network security- Mm-hmm ... and cloud security and identity- Mm-hmm ... and all the different, AppSec and everything else that we see today.
Yeah. And then all of a sudden, I guess it was what, maybe seven, eight years ago, now we call it EDR, right? Endpoint security, though, became, it's more than antivirus, it was more than the suite, it's endpoint security.
So what makes Bold? When you guys got together and said, "Okay, we've got to do something around endpoint security," what was the special sauce? What was the idea here that we're going to do something that no one else is doing?
Yeah. That's a great question. So, Mati, my co-founder, the CEO, and I, have been working together side by side for the past nine years.
The previous company- Mm-hmm ... was a DSPM solution that was got acquired in 2020, I want to say, something like that. Mm-hmm.
Maybe a bit later, 2022. And when we sold, positioned, pitched our cloud solution, one of the main use cases that customers really liked diving into was when someone is downloading sensitive information to the endpoint, and how we can see that and how we can control that. So that was a really cool part to present it and to show it and to monitor it through the cloud.
But customers kept asking, "Okay, so what happened next? " And being a cloud solution, we had no idea. But this question kept nagging, kept almost creeping up into every session that we had in the cloud space.
And this is when we understood that the focus is starting to shift, and people are looking into the endpoint more and more. And when we started Bold in 2024, there was really no debate that we saw that workforce is going into the endpoint back. People are downloading sensitive data, people are using sensitive data on their endpoint, and obviously, the AI usage brought a whole new boost and whole new light into that, because we see more and more people using local AI modules on their endpoint, adding more and more context to the agents that they're using locally, and that makes the endpoint probably the most critical frontier in the security today.
I don't disagree. So now that we've got Bold here, well, when did you found Bold, about three years ago you said? Two?
Two years ago in 2024. Yeah. So already AI, of course, is exploding on it, though it wasn't as much agentic as it is now.
It was more generative even then. Mm-hmm. But AI to me is a double-edged sword for security.
Mm-hmm. Right? Because in some ways it's a fantastic tool that we could use to do things at a scale we were not capable of before.
However, the bad guys can also use it as a tool to do things they were not capable of before as well, right? Definitely so. And so AI is both a sword and a shield- Mm-hmm ...
when it comes to security. How is Bold using AI? That's a very interesting question.
So first of all, I would say that we had the privilege, as you said, to completely rethink DLP with AI because we had even, I would say in a way, the opportunity, the luck, to build this company once AI was already out there and being used. So it's built into the core foundations of our company. The core solution of Bold is using language models on the endpoint for data classification, data lineage, tracking source to destination, obfuscation detection, anomaly detection, and we now added this really cool feature of actually an intent agent.
All of that is taking place on the endpoint. So imagine if you will, that each endpoint in the organization has its own individual agent that can take all this context into account and make a decision autonomously. So you have this very powerful tool that you can obviously command and control from afar, but it allows you to respond in real-time to threats without the need to do a cloud round trip to get some logs, to get a decision, to ask a bigger model if it's sensitive or not.
So the end user experience remains intact, uninterfered, unless there's a breach. And then to really scale it up so simply, because again, each agent has its own everything, context, decision mechanism, and protection mechanism. So we really built around AI and every new module that comes into light and that we could use on the endpoint is actually a really big step for us to take the product even further.
Sure. Yeah. Now, I wanted to jump in, and you've touched on it a little bit here, but when we're looking at internal traffic within the organization- Mm-hmm ...
distinguishing what is legitimate, and I don't even want to use the word legitimate employee. You can have legitimate agent traffic today. Mm-hmm.
Certainly, it doesn't necessarily mean it's a threat, so let's just phrase it this way. How does Bold, and using AI, how are you able to distinguish legitimate activity from insider threats, from malicious activity, from runaway agents? There's so many different things that can go on today.
So, a bit boring answer is context. We take into account pretty much everything. So that would include the user role- Mm-hmm ...
the sources of data, where the data came from, when was it last edited, who used this touch data in the company, across the company, where this data is going, all the way to the level of the logged account in these different services. So being on the endpoint, we're able to differentiate quite simply and easily between personal AI accounts and company AI accounts, as well as between different services. So even if your organization approves ChatGPT for everyone, I can distinguish between the different tiers of subscription levels and the different tiers of the logged account, and then to understand if this traffic is actually into a sanctioned application within the organization or alternatively, an unsanctioned one I guess the next question, though, is I get that you use context, right?
Mm-hmm. That's a good way of doing it, but it just seems so damn hard. I don't know a better way- It is ...
or an easier way of saying it to you. It is. It is.
Yep. It's just so hard. I look at some of the stuff that I see just in our network here.
Yeah. Right? And emails, and messages, and smishing, and everything else.
It's getting so hard to really distinguish. It could be quite difficult, but at the same time, having the experience that we have had its advantages. We've seen such communication within an organization before.
We have a very good understanding of how an organization is working. We are learning from our customers as well, how their specific organizations are working, because there's always this unique flow in the organization that you weren't familiar with. Something that might be risky to one organization is not risky to the other.
And we come with this premise that sensitive data is leaving the organization in order for the organization to do business, and that's okay. And I think a lot of DOPs look at it from a perspective of, "Oh, sensitive data has gone out. I need to stop everything.
" An organization cannot work without trading its sensitive data. It's all a matter of to who this data is being sent, what was the business purpose, does that make sense comparing to previous activity, the role of the user, its risk to the potential organization, and, if needed, to intervene in real time. Hadar, I'd like to use what time we have left because I didn't do it before.
How can people, all right, this is something that's interesting to them, how can they engage with Bold? What's the best on-ramp? security.
This is our website. You can see it all through there. I would be extremely happy to hear from anyone in the community, so just a DM in LinkedIn to my profile would also be excellent, or the company's profile.
But yeah, we're pretty much out there wherever we can, and we really like to have the opportunity to present to everyone how good is this product and what's coming up next. security. Is there a trial or something they could download to get their hands on?
Or is it you got to set it up with a salesperson? What do we do? So you do got to set it up.
Being an endpoint solution means that we have a very quick deployment, but a deployment nonetheless, on agents in the organization, on endpoints, sorry, in the organization. And while there is a virtual environment that you can test out the product and kick the tires and learn about it a little bit, there's nothing like seeing the product working in your environment, feeling how it behaves, how it looks, how it learns about your environment, and gives you very interesting insights about how data is being used within your organization. And building such a product and showing our customers, even during a very short POV, how much we learn about their organization and how much our agent has adapted to their risk tolerance and learned their sensitive data, there's just no second to that.
No. Agreed. security.
What about are you guys going to be at any conferences or anything coming up? Yes. Pretty much all of them.
So, we do so many of these events. Obviously, we have a big presence in the US, in West Europe as well. Yeah.
Go check us out wherever we have a booth. But yeah, it's just a great opportunity for us to show our product a bit more, to get the feedback from the community. We are really attentive to our customers, to our prospects, to learn from them, to learn from their issues.
The landscape is constantly changing. We do see it, like how the different profiles of risk have changed throughout the past just two years. So it's pretty much as you said, this AI thing.
Yeah. It's also an opportunity for all of us, but it's also rapidly changing. Exciting times, I would say, definitely.
Every day is an exciting time. Every day. Anyway, Hadar, thank you so much for coming up here on Techstrong TV.
Hope to see you back soon, and best of luck with Bold Security. security. Check it out.
You're watching Techstrong TV.