Akrites Rallies the Industry to Defend Open Source Security
Open source security has entered a new phase. AI tools are finding vulnerabilities faster than maintainers can fix them. The volume is turning a long-standing challenge into a crisis.
Sonatype co-founder and CTO Brian Fox examines that shift with Alan Shimel at KubeCon + CloudNativeCon in Salt Lake City.
Turned Up to 100
Sonatype runs Maven Central, where most of the world’s open source Java comes from. For years, it has preached knowing what is in your portfolio and being ready to upgrade.
Fox says those lessons still hold. The AI-fueled surge in vulnerabilities has simply turned everything up to 100.
Registries Are Not Free
Every patch produces new releases that must reach consumers. That puts pressure on registries that are already strained.
Infrastructure costs money, Fox notes, and much of the traffic is waste. Some organizations download the same unchanged JAR thousands of times a day. Maven Central now asks heavy commercial users to pay to publish.
Fox helped establish the Linux Foundation’s Sustaining Package Registries Working Group. Registries for Python, Ruby and other ecosystems share best practices there. Not-for-profit registries are especially exposed when donations dry up.
Grounding AI With Real-Time Data
Sonatype research shows that even frontier models are poor at recommending dependencies. Their knowledge stops at their training date.
Sonatype now feeds its data into models through MCP. Fox says a small model grounded in current data can beat the best frontier model.
A Front Door for Vulnerabilities
Financial firms with early access to AI discovery tools have found huge numbers of serious zero days. Flooding maintainers with redundant reports could overwhelm them.
That is the mission of Akrites, a Linux Foundation initiative where Fox is co-chair of the governing board. It ingests, validates and deduplicates reports before passing them to maintainers.
AI companies, hyperscalers and security vendors are all involved. Fox expects a bumpy 18 to 36 months as the industry pays down decades of open source security debt.
Transcript
Hey everyone, we're back here live at KubeCon. If you know this guy here, well if you know him, you know him. If you don't know him, you don't know him.
I guess that's sort of the logic of it. But he used to be a regular here on us every KubeCon, all through the year. Unfortunately, I don't know, we've missed you, Brian.
Yeah, it's been a few. It's been a while. Yeah.
But here's my friend Brian Fox. Brian is a co-founder and is it CTO? CTO of Sonatype.
Look, he has so many other hats, and we're going to talk about maybe his newest hat in a second. But Brian, welcome back, man. How are you?
Pretty good. Good. You know what?
If you don't mind, we'll spend a little bit of time foundationally, maybe people who don't know Sonatype. Sure. And then from there, what's new at Sonatype, and then I want to talk about your new position.
Okay, sure. So Sonatype, we've been around for almost 20 years now- Mm ... since we started the company.
We started out, we're the company that runs Maven Central, which is where all the world's open source Java comes from. I think we're going to touch on some of that. But we also provide tools and capabilities for organizations to better manage their open source portfolio, making better decisions about all those open source dependencies.
For more than 15 years we've been preaching all of the things, SBOM- Sure ... before it was SBOM. Understanding what's in your portfolio, making better decisions, being prepared to upgrade.
And I think with the AI fueled explosion, the vuln apocalypse that we're seeing now, all of the same things we said back then are still true. You still need to do those things, but now it's turned up to 100. Yeah.
I mean, look, I've been there every step of the way. Yeah. Right?
I followed Sonatype. We knew it was the right thing, but even I didn't predict that we would have to deal with all of this all at once so quickly. The scale.
Yeah. A lot of people, they get touchy when you say vuln apocalypse. I've been just saying post Methos.
But the fact of the matter is, I've been in security almost 30 years. StillSecure, the company I founded back in 2001, we had a vulnerability management solution then. And the fact of the matter is, we never fixed the vulnerabilities we found then.
We never did a good job. Those things get deprioritized. Yeah.
It was always robbing Peter to pay Paul, and I'm going to fix this one, but- That's right ... and now put a 10X or 100X on top of that and stuff's just going to break. Yeah.
I'm sorry. Yeah. And then on top of that, you guys have an awesome burden in that it's not just a couple of thousand nodes that I worry about scanning for vulnerabilities.
You are the repo, as you mentioned, for all of this Java and all of these code snippets and apps and code being downloaded. I don't know what the answer is to tell you the truth. I'm hoping maybe you have some.
We have some. Like I said, we run Maven Central, right? Right.
So when all of these new vulnerabilities get patched, those produce new releases that need to get to all of the consumers. We're in the middle of that, so making sure that our systems can handle a 10X explosion in the number of published events, the download events, all that kind of stuff is really important. Over the past couple of years, actually we started it back in Denver when we spoke.
Yeah. I was just starting to get on about we need to do something different economically for the repository, right? It doesn't make sense for one or a few companies to pay for all the distribution of open source.
I remember this. Yeah. That was sort of just a thing that was bugging me that day, and we've worked on it for the last couple of years.
In Maven Central, we've changed the model around a little bit, and we're getting organizations that are using this for heavy commercial use to start having to pay to publish commercial, because it's not just- Because it's a commercial source ... open source anymore, right? It's not open source.
" Correct. It's not free beer. Right.
And this is what I tell people, "Listen, as long as power is not infinite and free, shipping bits around the world over- Costs money ... " Infrastructure costs money. Right.
And this costs money, and we see lots of waste in the ecosystem. Most of it's not intentional. It's abuse by definition, but it's abuse by carelessness.
We're seeing organizations that download the same JAR that never changes 10,000 times every single day, right? Crazy. 9,999 of those are wasted.
It's crazy. And so, we've been on a journey to work through that, reducing the waste, figuring out models to help make it sustainable. We've created a working group within the Linux Foundation where we have the other registries- Really?
Ruby, Python, and- Oh, that's great. What is that called? It's the Sustaining Package Repositories Working Group.
Okay. And so we're getting- It rolls off the tongue. Yeah.
It sounds exactly like what it is. Very descriptive. And we meet, and we're sharing best practices and ideas because it's not just a Maven Central problem.
All the registries have this. Agreed. And not for profit ones, they have a zero sum budget.
They can't miss a quarter. Yeah. And so they're completely dependent upon handouts.
And when the economy goes up and down, those handouts dry up. That's when those dry up. Bandwidth doesn't, right?
And so, we're working to try to, we're leading the way for Maven Central to do this, but we're sharing some of these and encouraging the other repositories to do the same. Well, start developing best practices for the repos across the board. That's right.
And look, this is a problem in open source since the beginning of open source. Right. Yes, it's free.
Yeah, that's part of the allure, part of what it is, right? The code here. And I don't mean computer code.
I mean the DNA of open source. Yeah. But there's also sort of a, I used to think there was an understanding you can't take, take, and take- Yeah ...
without giving back something. Well, in the early days it was different, right? If I worked on Maven back in 2007, it was a laptop I already had with an internet connection I already had, and I was basically donating my time.
Right. That was the cost. But now, I think people would be horrified if they said, "Wait, you just built that on your personal machine?
" Absolutely. And so, okay, that needs to be built in a secure enclave in the cloud. Who pays for that?
That stuff's not free. It's donated by hyperscalers, but it's ultimately still not free. Yeah.
Somebody's paying the bill. Right. And these are, I think people don't think about the implications of what we expect of modern security, even for open source these days.
They don't have any idea. Right. I happen to know what it costs, this is going back, I don't know, four or five years, for instance, the Jenkins project.
Mm-hmm. What it costs to just maintain the Jenkins project. Sure.
I mean, I've heard- And that's six figures ... I've heard some large database open sourcings, it costs $60,000 in compute time to run the test every time they do a release. Yeah.
Oh, absolutely. Right? It's six figures a year.
Yeah. And here's the other beautiful thing. We're very lucky.
You and I have, our careers have coincided with the open source explosion, revolution. It has enabled, look around this floor. Right.
What is- Everything is built on it, yes. Yeah. Everything is built on it.
At some level, there has to be a little something giving back, right? Sure. Yeah.
It's a community. As I say, it cannot just take. That's right.
So I think we're making good progress on that. Good. That's been the thing.
I'm glad to hear that. This little AI thing has happened since we spoke last. A little bit.
A little bit. It's changed everything. A lot of the stuff we've been doing is still relevant.
We're building systems to help organizations with thousands of developers and tens of thousands of applications manage their open source portfolio. Now, you've got 10,000 agents doing the same thing, and so coding the rules about what's good, what's allowed, doing auto-update patches, is going to become critical. Yes.
So you have to have the infrastructure in place to do that. So I think we've been fortunate that a lot of the platform that we've been building all this time is perfectly aligned for this AI revolution, if you will. Yeah.
And it's interesting because we've done a lot of studies. As you know, we do the annual report on the open source supply chain security. And we looked at how well the models, even the frontier models are at recommending dependencies.
Turns out they're very bad, because they have a fixed point of view in time. They don't know new versions. They don't know new vulnerabilities.
Literally nobody would send their security team out to defend- They're as good as their last training. Right. Exactly.
" But that's what you're doing if you're just using models. And so what we do now is we're able to take our data, plug it into these models via MCP capabilities so that- Okay ... whether you're using the model to generate new code, you're using the model to assess existing code, it has the latest and greatest information about all these dependencies.
And some of the shocking findings are if you ground one of the tiny little models, the nano models, if you ground that with data, it provides better results than the best frontier model. No. I wouldn't recommend you write code that way, but it just shows that the importance is having the real-time context, not- Absolutely ...
what was trained into it. So I hosted a partner advisory and customer advisory thing yesterday for a large vendor. I'll leave it at that.
" Two people. " Four more people. Sure.
But that was six out of 36. Yeah. By and large, people seem to be relying on the hosted frontier model.
Sure. We've seen this transition before. Yeah.
Mainframe to PC- Sure ... cloud to hosted. We're in that transition again.
At some point, people will realize it's cheaper for certain use cases to use an open weight model on their own compute because they're essentially free tokens, right? Right. Just half of them are different.
So I think this reminds me of cloud. Are you going to be on a single cloud, multi-cloud? Yeah.
All of these things. Yeah. It's the same pendulum.
I think that's what we're going to see here. Yeah. You may not use that open weight for every single thing you do.
Right. But for 75% of it, you will, and you'll save a ton of money. Right.
For a fixed cost of buying a machine and racking it up, yeah, let it go. So yeah, that's a thing. And then we were chatting before about some of the vulnpocalypse stuff and what we started to see within many of our customers, we're strong in the finance firm.
Yeah. A lot of them had access to Glasswing early days. Yeah.
When they came out of their secret period, we started hearing, "Listen, we found 50, 80,000, 100,000 highs and criticals," not lows and mediums. They stop counting those. Yeah.
That are in the open source, and these are zero days. Right. " Upstream to everyone.
And then you've got 10, 20, 30, 100 companies doing the same thing. We know the maintainers are already overwhelmed with the amount of stuff coming in. If we bury them with redundant reports, low-quality reports- We're going to kill them ...
we're going to kill them. And so working with the Linux Foundation, a bunch of us have gotten together and created what's called the Project Akrites. Oh, okay.
This is what I wanted to talk about. Yes. Yeah.
And so the mission of Project Akrites is really to try to be that clearing house and that front door for all of these vulnerabilities coming. Okay. And so what- Across the board ...
Across the board. And so what we're trying to do is we're automatically ingesting these vulnerabilities, we're validating them, we're de-duplicating them, and we're creating better reports in the way the maintainers want them. Some projects, some foundations expect different things.
And so we're collecting all that so that when we get these in, we can clean them up and hand that easier to consume back to the maintainers. Absolutely. What a great thing.
Yeah. And so it's a- And this goes across all open source projects? All open source.
Which is- Not just CNCF. No, there's not a limit. And it's a big task.
What an undertaking that is. It is a big undertaking. We don't know the total number of actual zero days because everybody's still holding them embargoed until this process gets going, and then we can work them upstream.
So is it a million? I don't know. Who's working on this?
Oh, all the hyperscalers, the AI companies. They're all in. Other companies in the space, Chainguard, Herodevs.
Basically everybody touching the space is working together on it. That's a good thing. It is.
What's your role? I'm co-chair of the governing board. Good for you, Brian.
Yeah. You know what? It's funny, no matter how big this universe seems to swirl around us, there's always a few small people who volunteer and- We see problems, we like to solve them, right?
Right. And so it was immediately obvious to me, given all the stuff we've been doing for a long time- This had to be done ... but also our role with Maven Central is key in that, because we need to be tightly aligned.
Because when Project Akrites breaks the bottleneck of all these embargo things, and they start flowing down, a lot of them- A lot of them ... are going to come our way. Yeah.
And so we need to be able to work with that. But also, we're in a position where we have a lot of relationships with those maintainers already because they're publishing to Maven Central. So we can help bridge the gap.
And that's where a lot of the other people in the other ecosystems are doing the same thing. Plug it in. So it really is a little bit like a wartime effort.
Everybody has their own specialization, capabilities that they bring to the table. I think that's what we're going to take here. It's what it's going to take.
It's going to take that. So. Do me a favor.
If you could look into the camera, how do you spell Akrites so people could go find this? Sure. org.
One more time. org? org.
Yes. org. I'm sure you need more help.
We do. You need financial support. Right.
And as Brian mentioned, there's a lot of great companies already involved here, but this is a war when you think about it. It's a war. And we're going to need every hand, Rosie the Riveter and the rest of them, right?
That's right. And we know that the average time to exploit, we used to talk about days- It's 10 minutes ... it's not.
It's negative seven days- Really? Yeah ... on average, because these findings are being found and exploited by the bad guys before the patch, and even sometimes the disclosure gets upstream to the maintainer.
Just yesterday I saw the new version of Mythos came out yesterday. Yeah. Did you see this?
I did. 2, I think. Yeah.
Yeah. Yeah. We've got our hands full, man.
Right. And so it's going to be a bumpy 18 to 36 months, I think. Yeah.
Because what's really happening is we're paying down the security tech debt that's accumulated over decades. That's accumulated over decades. Yeah.
We spent 35 years. And we're kind of paying them one at a time. Out the other side, we might have much better software.
So that's what they say. But it could also be one of those situations where the operation was a success. Unfortunately, the patient died on the table.
I can't disprove that theory. Exactly. Yes.
Let's hope not. That's right. Anyway, Brian, always a pleasure, man.
Thanks for having me. Brian Fox. So it's Project Akrites.
Spell it one more time. A-K- A-K-R-I-T-E-S. The Byzantine Defenders of the Western Frontier is what that means.
Excellent. org plus Sonatype, Maven Central. Go check it all out.
We're live. We're at KubeCon. We'll be back.