Josh chats with Patrick Garrity about the VulnCheck State of Exploitation 1H-2026 report. Patrick explains the current trends we are seeing around vulnerabilities right now. While the number of CVEs is way up, the number of actually exploited vulnerabilities isn’t growing year over year. This tells us there is a lot of FUD and hype. We also ask where are all the vulnerabilities that project Glasswing found. They should be going public by now, but we’re not seeing that play out in the data.

This episode is also available as a podcast, search for “Open Source Security” on your favorite podcast player.

Episode Transcript

Josh Bressers (00:00) Today, open source security is talking to Patrick Garrity, a security researcher at VulnCheck a person I I love talking to. Patrick’s a good friend, so this is a ton of fun. And you just put out a report titled VulnCheck State of Exploitation 1H 2026, which is a very catchy title I will commend you on. But all right, Patrick, tell us who you are.

Josh Bressers (00:22) And and tell us why you’re here and we’ll go from there. Because I this report, like I have

not been this excited about a vulnerability report in a long time because like I’m gonna ruin it for everyone. It’s all bullshit, folks. That’s like that’s the title.

Patrick (00:35) Yeah, so thanks for having me on. it’s always a pleasure, always hanging out with you. and yeah, I d you know, the I think the major thing is we’ve seen, you know, that this first half exploitation, I do these every six months, right? And I’ve been

Josh Bressers (00:50) Yeah, yeah.

Patrick (00:51) doing them for like probably two and a half years since I joined VulnCheck Maybe two years.

And I just look at exploited vulnerabilities and try and understand trends and see what products are getting hit and you know see what the timeline to exploitation is. So like I have a pretty good sense of like consistency. You know, there’s always imperfections in vulnerable data, but generally speaking, it’s like the same methodology. Let’s look at some different sets of data in different areas. And I do an annual and then I do a half-year report. And the the really interesting thing for this half-year report is

the amount of of just insanity related to CVE and vulnerability disclosure volume. And then the the other side is like this narrative that AI is going to end the world tomorrow. I think

Josh Bressers (01:41) Yeah, yeah, yeah.

Patrick (01:42) quite as like the the vulnpocalypse and so I

Josh Bressers (01:46) Yes.

Patrick (01:46) just I just I’m like, hey, I have some questions. I want to take a pragmatic view to this and and try and understand if like the data is really

starting to speak to like the fear, uncertainty, and doubt that these frontier models are are promoting. and and that’s

Josh Bressers (02:04) Yes.

Patrick (02:05) I mean that’s more or less, you know, the the the questions I think I put at the top are like that I was looking for are you know that I was curious about are more vulnerabilities being discovered and disclosed resulting in more vulnerabilities being exploited? Is the rate at which vulnerabilities are being exploited faster?

Or you know, because there’s this narrative that like, AI models are accelerating exploitation faster, quicker. are vulnerabilities being discovered with AI tools more dangerous? Or are we all feeding into the AI-assisted vulnerability discovery hype cycle? and and you know, in in I did some research the whole time that that this narrative has been played out.

and so yeah, it’s just been a lot of fun to like, you know, kind of be on the sidelines and collecting intel and doing vuln disclosure and just watching the ecosystem just like I don’t know, freak out. it’s

Josh Bressers (03:06) Yeah, yeah.

Patrick (03:07) it’s it’s you know, you gotta be a little bit cynical, but somewhat entertaining. But then also like there there’s a reality that defenders really have to understand what’s going on. and so

Josh Bressers (03:20) Yes.

Patrick (03:20) that’s more or less

These are real challenges people ask me all the time and I’m like, let’s let’s go into the data and see what it tells us. So

Josh Bressers (03:27) Yes, yes.

Which I love. And I absolutely adore any time we can tell a story with data. That’s like my favorite thing in the world. So I want to start with one question because you talk about KEV which stands for known exploited vulnerabilities. And we usually think of KEV there’s something called the CISA KEV which I think is what a lot of people think of. And your list of known exploited vulnerabilities is not the CISA list of known exploited vulnerabilities. So why don’t you kind of explain the difference

Patrick (03:52) Yeah.

Josh Bressers (03:53) to that for us? Because I think that’s important for the rest of the discussion.

Patrick (03:56) Yeah, yeah. So it’s inclusive of CISA KEV So everything on CISA KEV we have our we we have a free product called VulnCheck KEV We give it away and we we essentially collect and curate any anywhere that’s public information, where there’s known exploited vulnerability evidence of exploitation of a vuln that could be anything from like a honeypot being hit via Shadow Server or CrowdSec or KEV Intel we also have a network of

a real vulnerable host called VulnCheck Canaries and we put indicators that those have been exploited. and and then like you have a lot of disclosures via advisories. So organizations like Microsoft or Google or Apple. I actually have a chart on the bottom I’m trying to like patch stack in the WordPress ecosystem and WordFence are commonly disclosing exploitation, which is pretty common. BitSite, Cisco,

a bunch of Intel organizations, like mm exploitation often comes from like just a broad set of sources. And so the objective and goal of of our team and with Kev is to like curate here’s all the things where people have talked about them being exploited. doesn’t mean they’re all equal. Like some some of the technology might be like a nation state and get hit hits once and some of it might be like broad spread spread

a strain from a botnet that has compromised, you know, thousands or or hundreds of thousands of hosts. So it can vary, but that that allows us to look at like, hey, this is what what is known. Like it’s imperfect, but but it’s a pretty good sense of these are things that are bad and you probably should fix them if they’re in your your environment.

Josh Bressers (05:45) And your

list is larger than the CISA list.

Patrick (05:48) Yeah, just past 5,000 KEVs So it’s it’s about I don’t know, two or three times as large in size as far as number of known exploited vulnerabilities. CISA tends to be like very, very focused on broad enterprise technologies. So like you’ll you’ll find other technologies like end-of-life devices a lot of times that we’re seeing exploitation or Chinese manufactured devices that

maybe might not be in the US ecosystem or they might be and people just

Josh Bressers (06:22) totally

Patrick (06:22) don’t realize it. embedded products are real. you know, so the there’s you know, also like they’re they don’t track WordPress plugins really, unless it’s like extremely broadly deployed. so there’s just a, you know, a lot of difference from an evidence perspective of of there’s more signal out there.

and then a lot of times we have it much earlier, because our criteria’s all you know, we see it at it. you know, sometimes too researchers they they’re like, hey, we see this, and they put it on X or they put it on LinkedIn, and it’s hey, yeah, that’s a reputable person. Like, let’s trust their within reason. Like, we’re only talking about like two percent of all vulnerabilities. So

Josh Bressers (07:08) Yes.

Patrick (07:08) 5,000 is relatively a small set of things.

you know, worth prioritizing, fixing and i you know, if they’re in your environment. So that’s what VulnCheck have is.

Josh Bressers (07:20) So I I want to talk about that. You you just mentioned what five percent of vulnerabilities. I’m trying to find the graph. Here it is.

Patrick (07:26) like

two no, five thousand so like two percent or less less than two percent at this point.

Josh Bressers (07:28) Okay. right. But but so

this is this is the story, I think, from my perspective. Is if you look at the number of CVEs being published right now, it’s like completely out of control. And there’s like there’s almost this cult that is developed around like obsessing over the number of vulnerabilities being published. And it is a lot. I’m not gonna say it isn’t. But when

Patrick (07:47) Yeah, yeah. Yeah.

Josh Bressers (07:51) you we look at your data of the graph of the number of exploited vulnerabilities, which

theoretically are the ones we should care about and the rest I won’t say we shouldn’t, but we we probably won’t ‘cause we don’t have the time. But like that number is actually decreasing, it looks like, over time.

Patrick (08:09) Well, the total numbers incre it was up ten percent over like the half year, compared to last half year and it’s up mildly, but like CVEs are up forty five percent this year. I think on the first half of the year. Exploitation’s up like five to ten percent. and and so like, yeah, not you know, that means that ratio is dropping. Now, at the same time, like those aren’t a cohort analysis, so it’s not one to one.

necessarily of like these ones that got published also are being exploited. So it’ll be interesting to see like how it draws out over time. but generally speaking even when I do cohort analysis like the cohort analysis is showing and that like the same line of things are being exploited you know generally count-wise over time.

there’s definitely been increases over the year, probably more related to visibility. and then I think like, yeah, attackers shifted in twenty twenty-one from credential compromise to more exploitation, likely because multi-factor authentication hardened the ability to do credential compromise. but yeah that I mean that’s generally it it’s showing like exploitation is is

persistent. And I think there’s also a question of like, is it that the threat actors just have so much like there’s only a finite number of people that are are doing bad. Like maybe it’s growing AI makes it more accessible, like which is true. but generally you know they they’re they’re they know the things they want to target and they exploit them over and over again. If you look at the products that continue to get hit. definitely if a product is is broadly used, like it’s going to be

likely to be targeted you know more frequently and then you could also talk about like is the code is the code base that good if it’s continuing just constantly to get popped as well so there’s a lot like there’s a lot to unpack there but yeah I think I think you’re right it’s like do all these vulnerabilities matter? Are all these vulnerabilities in the the enterprise ecosystem? Like those are those you know those are great questions.

And I think, you know, if you look at it, it’s like, yeah, just the WordPress ecosystem’s kicking off like gosh, this year maybe like five to ten thousand CVEs alone. So

Josh Bressers (10:37) Yes.

Patrick (10:38) if you just stay away from WordPress plugins, like we’re not even talking about WordPress Core, which is generally pretty hardened, although recently we have a new yeah, that it was you

Josh Bressers (10:47) Except what, last week.

Patrick (10:50) know, but but s also content management systems like WordPress and Joomla and and I think Oracle.

Maybe WebLogic might be in that category, maybe not. they tend to be targeted because they’re on the internet and they’re everywhere.

Josh Bressers (11:04) Yeah, right, right.

Patrick (11:06) so like generally those are just technologies that like you want to patch and you want to patch fast and religiously and also have mitigating controls like a WAF in front of them. you know, web application firewall.

And so yeah, the I mean I think the volumes like you want to compartmentalize it down into like what are the actual technologies being targeted is is a thing I like to do. you know, from a KEV perspective. So then I can understand like where I spend my effort and and automating, building mitigating controls, like nothing I I don’t think things have substantially changed as far as what

the what controls you should be putting putting in place. I think the speed in which you might be remediating should probably be faster. I think that’s always been the case though. as well.

Josh Bressers (11:58) So I am I remembering

incorrectly though, the the time to exploit hasn’t like significantly dropped since last year. In fact, didn’t it go up a little bit since twenty twenty five?

Patrick (12:09) well, kind of twofold, right? So in the first half, like overall it seems like there’s less vulnerabilities being exploited at the time of disclosure or the same day of disclosure or before. so that tells me a little bit and really like it’s not that far off. I think it was twenty-nine percent last year. Now it’s at twenty-three percent of all Kevs. and it kind of shifts. So like it’s in a twenty twenty-three to twenty

30% range usually. so that that’s consistent. But the thing that changed was the the median time, like the middle of of the road of all the vulnerabilities. Like that was 120 days last year. The first half of this year it’s at 80 days. So it does

Josh Bressers (12:55) Okay, okay.

Patrick (12:56) tell me that like people are people either are moving faster or we have earlier detection, right?

Josh Bressers (13:05) right.

Patrick (13:05) And my my

gut is like they’re probably moving faster, but also like it maybe there’s better detection capabilities that this stuff is surfacing quicker, faster on the exploitation side as well. You know, there’s always

Josh Bressers (13:20) Right, right.

Patrick (13:21) different dynamics that you could consider looking at that are the cause of that.

Josh Bressers (13:25) Well, I mean it’s the old c

correlation isn’t causation, right? I mean, that’s always the challenge

Patrick (13:30) Yeah. Yeah.

Josh Bressers (13:31) with this stuff.

Patrick (13:32) Yeah. Yeah. And like you can you you can spend so much time going down each of these rabbit holes of like, well, is it detection is better? Is it threat actors are moving faster? is you know, I think the the narrative of like, AI is all sudden, you know, that that’s probably the big part we’re gonna talk about is like is AI suddenly shifting everything to like urgent, immediate, the world is ending.

you know, and that and so that’s a big part I looked at was twofold. Like, okay, of the AI discovered vulnerabilities, like, you know, I think that’s an interesting data set of

Josh Bressers (14:11) Yes.

Patrick (14:11) like all these frontier models and different projects discovering vulns and disclosing them. Well, some of them disclosing them. like, are those being

Josh Bressers (14:22) Right.

Patrick (14:23) exploited and what products are they in? And then I think the other side of that is like.

actual AI products being exploited. and I think we’re seeing a lot more news about some of these products you know, that are are getting targeted and hit by attackers. And and you know, kind of we could go down each of those paths, but like I think actually the risk of AI weaponization of vulnerabilities, like an AI model finding, discovering, and doing exploitation, like to me.

I’m I’m not the the data’s not really showing, you know, that that there’s a significant change and that’s causing disruption. On the other hand, there is a significant change in the number of known exploited vulnerabilities in AI products. and so what

Josh Bressers (15:14) Yes.

Patrick (15:14) that means is like, yeah, we just expanded our attack surface, like the the products that you’re implementing within your environment that are related to AI,

Number one, they’re new products. A lot of them are not hardened. They have a lot of vulnerabilities. And yeah,

Josh Bressers (15:31) None of them are hardened, Patrick. None of them. Literally none.

Patrick (15:36) they’re all vibe coded. but yeah, to to me that like that’s the bigger exposure from an enterprise perspective is is

Josh Bressers (15:45) Yeah, yeah.

Patrick (15:46) you know that the AI products themselves. that

Josh Bressers (15:51) So I d I’m I’m

trying to look. I don’t see anything in here that that filters some of this up. When we’re talking about AI products, you mean it’s everything from like open claw agents to people running like claude code on their desktop, right? That that’s the kind of things we’re talking about.

Patrick (16:06) Yeah,

yeah, so so a lot of the connectors, I kind of listed this out and and yeah, like you know a lot of things that you would connect with claude code or open AI, a combination of like AI gateways, AI agents, AI workflow automation, workload scaling, model building. and a lot of these tend to be open source products.

Right. So some of some of the products is example lang flow light lm open claw n8 n8n I probably said that wrong. Olama Great Gradio Comfy FlowWise Flowwise is getting hit. There’s a ton of new KEVs this month that aren’t even in this report for Flowwise, I think.

Josh Bressers (16:59) interesting.

Patrick (17:01) Yeah, so

the the the thing that we’re observing because we have canaries with a lot of these technologies. we see both first stage and second stage and and what we’re seeing is they’re doing credential harvesting and then they’re also running crypto miners. which is

Josh Bressers (17:20) Yeah. Yeah.

Patrick (17:21) n the same the same thing like attackers have always been doing mass compromise wise. But the credential side gets really interesting because now

You know, okay, am I leveraging that to use the model? Am I using that to harvest data from your your instance? a lot of these tools also are interconnected with you know your core infrastructure because people are running a hundred miles an hour at implementing AI.

Josh Bressers (17:48) Yes.

Patrick (17:49) and so keys to the kingdom, like full privileged act you know, full privileged access to anything that you want.

type of scenarios. and so yeah, you know, from my perspective, it’s like, yeah, that the that this seems like the bigger risk here. more so than like the frontier models and and you know glas glasswing mythos GPT eight point ten nine seven. I’m thinking I’m not I can’t I

Josh Bressers (18:19) I know, I can’t keep track either.

Patrick (18:20) I can’t keep track. but yeah that I mean that that’s yeah

Josh Bressers (18:23) Okay, I w I wanna I wanna ask about that though. You have a graph

in here. I keep scrolling up and down in this. I should have gotten a better scrolling device prior to this this discussion. But you you have a graph that there you go. I should get that would no, not right now. All right.

Patrick (18:33) You need you need one these like Apple Pad things. Yeah. Very nice.

Josh Bressers (18:41) You you have a graph that shows the number of disclosed vulnerabilities from Glasswing versus the number of things that should have been disclosed due to the ninety-day disclosure period.

And it’s it’s a pathetic graph, right? Like the number of things that

Patrick (18:57) Yeah.

Josh Bressers (18:57) should have been disclosed is what? In thousand, more than a thousand, fifteen hundred or something. And the number of things that have been disclosed is is what is the number? Like twenty seven, I think I see.

Patrick (19:07) Yeah. Well,

there’s more that have been disclosed that I’ve been able to find. They just don’t update their ledger. And so I think like pulling a little bit back here, like, you know, gl f I think most people know what glass swing mythos is. Like, hey, we have this new frontier model. It’s too dangerous for the world. We can’t give it to anybody. government, you should ban models like this. wow, we got banned. What? okay. But

They came out and they were like, we’re f discovering thousands of vulnerabilities and in the in the 400 page re of reports and their cards and their models and all this. Like I I honestly think like these people are like the people that are involved in this, like they’re very smart. But also I think there’s a lot of manipulation going on in relation to like,

Josh Bressers (19:52) Yes.

Patrick (19:53) let’s publish so much volume of data that no one can consume it that you’re literally just going to be like, whoa, this is world ending, right?

The whole report out of like 250 pages and all the information with Glasswing had a single vulnerability disclosed. That same exact vulnerability was actually discovered by Aisle another AI project, pretty cool team,

Josh Bressers (20:18) Yeah, yeah.

Patrick (20:19) that found the vulnerability and reported it and also provided the patch for it well before Glasswinger reported it. So for me, I’m like, something is off here.

Like something is off. Like if you’re gonna come out swinging and say, hey, we discovered thousands of vulnerabilities, provide evidence with it. Like

Josh Bressers (20:36) Yes.

Patrick (20:36) provide evidence that’s substantial. And so for me, it was they’re like, we’re gonna issue this ledger and you’re gonna be able to track every vulnerability. It’s gonna have hashes. I mean, all the technical details are there on how this is going to be done. And then they launched the ledger like in May. And at this point, I’m tracking to see, you know, manually.

Like to to track for myself because I’m like, dude, I I don’t I don’t trust these. I don’t trust this. Like

Josh Bressers (21:05) Yeah.

Patrick (21:05) I have gone I I read so many vulnerability reports and I’ve seen so many things and I’m like something is off in relation to the approach that anthropic is taking to this. Like transparency, disclosures, like so many things are missing. But but they caught everyone’s attention and freaked them out.

Josh Bressers (21:26) yeah, yeah.

Patrick (21:27) And so I ended up collecting manually off CVE records, advisories, a bunch of different places, like where anthropic was attributed to. And and you know, there it like I’m not dis disputing that the that Claude and and OpenAI and these models can discover vulnerabilities. like we all know that’s true. We’re using and leveraging the models, like we it with the right with the right knowledge and understanding, you can get it to do wild things and find things and things that you can exploit.

and and and I think what that’s really re seen in the CVE volume that we’re seeing is like vulnerability

Josh Bressers (22:03) Yes.

Patrick (22:03) discovery generally has become accessible to the masses like it has never been before with AI models. and and so naturally it’s like okay, like cool, I think we understand that. But the the the the piece with anthropic is is

Then they’re like, we have this ledger. They released the ledger. And I at this point I have like 85 vulnerabilities that I’ve been able to attribute to them based on like t tags and CVE records and advisories and credits and all that stuff. And the ledger had 27 vulnerabilities that were disclosed. And they’re like, we have a 90-day window. Like our VDP is that in 90 days we’ll disclose, or we’ll disclose once a patch is available. But so I’m like, okay, so like I already have 88 things that are disclosed and your ledger only has 27.

and s and there were a few on the ledger that I didn’t have yet, so I pop those over, right? Now I have like a hundred. I don’t know. and I’m like, there’s more, they’re not maintaining this ledger. I’m gonna build a tracker so I can track all this stuff so that it will notify me when it updates. I mean, a month

Josh Bressers (23:08) Nice.

Patrick (23:09) goes by and a single update on the ledger doesn’t happen.

Josh Bressers (23:12) Whoa.

Patrick (23:13) Like

And and so I’m like, well, this is interesting because like their 90-day disclosure policy, all these vulnerabilities. I think at this point it’s like two or three hundred vulnerabilities have passed the 90-day window. There hasn’t been a single update. So like they just abandoned the whole project. They just they they didn’t

Josh Bressers (23:28) I I I I I bet they did.

Patrick (23:31) abandon the like my opinion is well, and I think we’ll there’s a combination of things going on here where like yeah, they they well and when they came out of the ledger, they’re like, We found 23,000 vulnerabilities.

Josh Bressers (23:43) Yeah. Yeah.

Patrick (23:44) Or find they call them findings. And then they were like, we validated like 500 of them and reported them, but they had to hire four or five different companies to do validation before reporting these vulnerabilities. So, you know, part of me is like, did they just pick the ones that like were obviously real vulnerabilities out of this 23,000 data set and and then ran all the manual work to validate them and get other people to report them for them?

But to me, what that shows is like the bottleneck in vulnerability discovery. There’s a huge bottleneck in related in relation to vulnerability coordination that I don’t think any of these people that are running these projects

Josh Bressers (24:25) No. No.

Patrick (24:27) understand or know. And like it’s like they’re they’re like surprised, like, man, I report vulnerabilities and nobody at the project, an open source project listens to me. Like

They actually need and and they even they even signaled at one point, I think it was on a podcast or something, that like, dude, some some projects have asked us to stop reporting vulnerabilities because they’re overwhelmed because they can’t manage the intake.

Josh Bressers (24:51) Yeah, there’s some of that for sure.

Patrick (24:53) Yeah, and so I I think like a combination of things going on here and like that ledger, it we’re past two months from when it was released. It has just been completely abandoned. from day one.

Josh Bressers (25:05) Yeah. So I I have a theory, Patrick.

I have a theory. And I will tell you my theory, and I have a suspicion you will not disagree with me, is that

Patrick (25:08) Yeah yeah. Yeah, what’s your theory? Yeah. Yeah.

Josh Bressers (25:14) they probably did find a a absolute bucket full of problems. And I’m sure

Patrick (25:19) Yep.

Josh Bressers (25:19) some of them were valid, some of them weren’t. And I have a suspicion once they started reporting things and realized that the the we coordinated vulnerability disclosure, that’s you know, C V D, that’s the term we use in this universe. They figured out that

That coordinating this stuff is not just a matter of like sending a bot loose on a GitHub repo, right? Like this is a very human-focused task where you have to, there’s

Patrick (25:43) Yes.

Josh Bressers (25:43) always always back and forth. And and I have received an enormous number of AI written vulnerability reports, and they are like rage-inducing because it’s like this should have been four lines of text. Instead, I have 23

Patrick (25:54) That’s four pages long. Yeah.

Josh Bressers (25:58) pages I have to read to even figure out what the heck is going on. My guess.

Patrick (26:03) Yep.

Josh Bressers (26:04) Is they learned it’s hard and they’re gonna just pretend they never did it and hope no one notices because but that doesn’t mean they probably are sitting on thousands of vulnerabilities somewhere in some internal git repo. Like that’s the thing, right?

Patrick (26:16) Yes. Yeah, yeah, yeah, yeah. Yeah. Yeah.

Well, that’s the other piece I’m like, it’s like, well, now you’re just sitting on all this stuff. And by the time you at by the time you get to it, it’s already reported by other people because these AI models are people independently

Josh Bressers (26:30) Yeah. Yeah.

Patrick (26:31) with AI models are finding the same exact things. And that’s what the data showed me was like, as I went through it and saw the things that they were reporting, other people had already reported them.

in in in in maybe some new discoveries and maybe they’re like those same people are using claude to find it. I’m not disputing that, but yeah, I just think genuinely kind of interesting. Then then I think beyond that you start to see and we’re we’re going a little bit outside the report, but I think like that’s what’s fun is like the navigation of all these like different poles

you you see industry projects of people coming together suddenly to to do vulnerability clearing houses and do coordinated vuln disclosure. And I think it’s gonna be fascinating. You have the White House with Golden Eagle, you have a Akrites which is Open AI, Claude, and all these AI projects, and then you still have like thousands of independent researchers as well. and and it’s gonna be interesting to see if there’s what comes out of these projects.

Because I it’s really easy to get together in a group of people, and you know this. You’ve been an open source a long time, be like, hey, let’s all get together and do this thing. Everyone shows up to like shake hands and be like, yes, we agree, let’s do this. Execution is completely different. Like execution and you you you

Josh Bressers (27:54) Yeah. And this is a super hard problem.

Patrick (27:58) Yes, yeah. So like I don’t want to discourage it, but at the same time it’s like, yeah.

And I’m happy to like I’ve offered, I’m like, hey, happy to participate. Like we actually run C V D for like hundreds on a monthly basis of of vulnerability disclosures. Like go look at our advisories. We shoot 5,000 CVEs so far in the last like I think we just hit 5,000. But in

Josh Bressers (28:22) Nice.

Patrick (28:22) but like 4,900 of those were in the last like 12 months. like

Josh Bressers (28:27) yeah, yeah.

Patrick (28:28) we’ve done we’ve done volume to understand like what are the challenges, what are the problems, how do you how you might approach this.

you know, generally when I reach out to these projects, I get crickets. And so I’m like, yeah, like I don’t know that they’re I’m seeing real interest in actually solving the problem. I think people are like, We’re gonna own this, we’re gonna we’re gonna do it with with all the intention in the world. but

Josh Bressers (28:57) Yeah.

Patrick (28:57) I’m I’m really curious to see what comes out of the other side on this, because like

By the time all these vulnerabilities get to this place where like now we have resources to do coordination, which requires humans, you know, are they are they gonna like fund hiring the humans to go do all that work? are you know, like generally the people most of the people I see on these projects are like executives or leaders, you know? So and I’m

Josh Bressers (29:21) Right. Yeah. Yeah. People who go to meetings.

Patrick (29:23) I I’m not yeah, I’m not trying to criticize, I’m just being like sitting in reality. It’s like

Yeah, it’s gonna be really interesting to see how these different projects turn out.

Josh Bressers (29:34) Yeah, yeah.

Patrick (29:35) and I would I like number one thing, I love collecting phones, I love volume data. Like I I like with with anthropic, I was like, I just wanna see all the disclosures. Like they even were like, hey, if we don’t if we don’t get acknowledgement, it doesn’t get fixed after 90 days we’re gonna disclose it. Nothing, right? and so genuinely this has turned out to be like

Pretty much what I expected. on day one of Glasswing. Like day one of Glasswing, like almost, almost to a T. Like in my mind, I was like one vulnerability in the whole report. Like, I know how this is going to play out. so it it’ll it you know, it’ll be interesting. I don’t want to discourage people on these different projects and the clearinghouse and all that stuff. Like I I, you know, I wish them luck and I hope they’re successful because I think like that would be great.

I just yeah, I mean, I like you’ve been at this long enough. Like I have a huge amount of skepticism of like whether whether any of these projects will be w able to execute in reality. So yeah.

Josh Bressers (30:40) Yeah, yeah, for sure. I mean

it it is it is an industry that makes you bitter is almost what I would say. Like because unfortunately I think we have seen so many instances where it you’re promised countless things and it never happens. And it’s just it’s it’s it’s like

Patrick (31:00) Yeah.

Josh Bressers (31:00) constant frustration. I feel like the number of times I can count like a good thing happened in the world of vulnerabilities is over the last like

Twenty plus years I’ve been doing this is like three, you know.

Patrick (31:09) Yeah. Well I think I

think I like I think there’s a lot of like awareness and a lot of interest now and actually

Josh Bressers (31:17) Yes.

Patrick (31:18) like doing some stuff. So like that’s that’s interesting. but yeah having participated in in CVD program and in some of these other you know programs, it’s like they move very, very slow.

And it it it it the

Josh Bressers (31:34) Yes.

Patrick (31:34) the the actual the there’s a lot of interest in making things better, but there’s not what I’ve seen is there’s not a whole lot of action and following through with that and there’s not a lot of accountability.

Josh Bressers (31:47) Yeah. There’s no accounting.

Patrick (31:50) and so you know I it we’ll we’ll we’ll see. but it yeah, it is kind of interesting when the government like has a has has like cert and they have the CVE program, they have CISA.

Like they already have these functions. Like why why

Josh Bressers (32:04) Yes.

Patrick (32:04) why are we spinning up yet another entity? but ul ultimately, like, yeah, more vuln data in different places. Like, we’ll go collect it for everybody and put it in one place. yes,

Josh Bressers (32:15) The X K C D cartoon, Patrick. That’s

Patrick (32:18) yeah. Yes. So it’s it’s it’s a fun one. yeah.

Josh Bressers (32:24) Yeah, yeah. All right. All right, man. Let’s let’s

take this one home. So what like what what is the takeaway I should I should get from this report other than my we’ll say it’s all bullshit that I led with? Cause I feel like this is a good report and there’s some really cool data in

Patrick (32:36) Yeah, yeah. Yeah. Yeah.

Josh Bressers (32:40) this report.

Patrick (32:41) Yeah. I I mean the biggest takeaways for me are like, yeah, the AI using AI is definitely exposure that hasn’t been there before. and so you know, if I’m an enterprise, it’s like you you really have to take this seriously that people are leveraging and building with AI, they’re going to, you want to enable them, but like this is a whole new attack surface.

and so that, you know, that’s probably like the the number one thing. I I do think that like, you know, I’d I’d be a little bit careful that that there is a large fear, uncertainty, a doubt narrative right now. And I’m a little bit concerned that it it we’re seeing it drive federal regulation, right?

Josh Bressers (33:22) Yes.

Patrick (33:23) and so I I think it’s great that like, you know, I think this week the the open mod you know, the open model aspect of all these companies being like, no, no, US government, please don’t build

ban you know open source models metrics I don’t I don’t know all the AI there’s so many AI terms that are like

Josh Bressers (33:41) Yeah, yeah.

Patrick (33:43) but but open weight that’s what what I was gonna say open weight models yeah I think that to me those are probably the biggest things is like we we shouldn’t overreact. I I do think that there is there is a large narrative being played to try and limit competition.

you know as well and naturally you know fear fear can drive that so i think you know advocate for open source advocate for open models i i do think be super mindful of open source it’s like yeah if you can submit a patch and contribute to the project rather than only reporting vulnerabilities that can help maintainers and and be really understanding that like

not every open source maintainer it’s a priority to maintain their project. I think that’s you know that’s a reality of open source. you know and then I think if you’re build the other side I would say is like if you’re building products, like everyone has their own products, every company, right? Whether it’s a website or whether it’s building software internally or externally or for customers.

but yeah, the the AI vuln discovery is real. So like, you know, get educated, get a harness, run different models against your code to see what you can find. Cause the other side of the fence is like on the coordinated vuln disclosure is yeah, if you don’t fix your stuff, like people are gonna come at you over and over and over again. and so there is a aspect of capacity here where like

You’re going to be better off if you actually go and look and take care of your own vulnerabilities, opposed to letting third parties. And the other thing within the data that I think was pretty validating as well is AI discovered vulnerabilities were both in commercial and open source projects. So it, you know, I think open source projects are like an early indicator because they’re an easy target.

but these these vulnerabil these AI models are discovering vulnerabilities in commercial products where maybe the code isn’t available. so just because like just because your code is not open source doesn’t mean that someone’s not gonna find the vulnerabilities on in it with these AI models. so yeah, hopefully that’s I don’t know, is that a good good summary of things for people to take into consideration?

Josh Bressers (36:10) Yes, I think it’s great. And I really like when you led with like don’t be afraid because you’re right, there is an enormous amount of FUD right now around vulnerabilities. And I think your data shows it is truly FUD because it’s like it feels soul crushing and like world ending, but it’s really not when you look at your numbers.

Patrick (36:30) Yeah.

Yeah. And then I I do think like it’s you know, threat intelligence is helpful as there’s more vulns because, you know, ultimately it’s gonna point you in the right direction of like the wor worst things to fix. and then yeah, anything with external exposure or user interact, you know, users interacting with like patch that stuff quick, fast. because naturally like network edge devices, content management systems, internet exposed devices are are right.

from a threat actor perspective for targeting. And and when you when you look at it, it’s like, yeah, a lot of these, you know, a lot of these organizations are spitting out a ton more vulnerabilities. So like, yeah, if you patch your Windows systems and you patch your firewalls and your network edge devices and this stuff in a timely manner, ultimately that’s, you know, really the the best thing that you can do. And you know, and then mitigating controls.

Josh Bressers (37:27) Thank you, my friend. It has been a treat as always to chat. I love it.

Patrick (37:32) Fun. Thanks so much.