AI coding assistants pull packages into codebases faster than security teams can track or patch them. Every package brings dependencies and transitive dependencies that ship to production without a record of where they came from. Open source is most of what teams ship, and malicious packages keep arriving through npm and PyPI while scanning alone fails to catch them.
In this interview on TFiR, Abby Kearns, CEO at ActiveState, breaks down how AI coding assistants multiply untracked dependencies, what EU CRA means for software bills of materials, and why teams should start secure instead of patching later.
Guest: Abby Kearns, CEO at ActiveState
Show: The Agentic Enterprise
LinkedIn: Abby Kearns
Watch the discussion on YouTube
Here is what every security team and platform engineer needs to know.
Technical Deep Dive
Q: Why do AI coding assistants create software supply chain security risk?
Abby Kearns, CEO at ActiveState, says every language, whether Python, Go, Rust, Java or .NET, brings packages, dependencies and transitive dependencies that attackers now exploit faster, with new npm and PyPI incidents reported every week. AI coding assistants make this worse because they pull in packages from everywhere as needed to solve a problem, and none of it is tracked before it ships to production.
“It’s not being tracked; it’s all being shipped to production.” Abby Kearns, CEO, ActiveState
Q: How do you track the software supply chain and SBOM when AI writes the code?
Kearns says the industry swept software bills of materials and supply chain visibility under the rug for years, and AI now adds even more packages that no one tracks. Organizations struggle to see the breadth of their supply chain and the provenance of what they pull in. The answer is a secure place for developers to pull from, so they are not pulling packages directly from the package managers.
“Organizations need to become more aware of their entire software supply chain from a security standpoint.” Abby Kearns, CEO, ActiveState
Q: What does EU CRA Phase 2 require for software bills of materials?
Kearns says CRA Phase 1 went into effect this past September and sets a timer on notification of exploitation. Phase 2 arrives in December 2027 and requires organizations to understand their full software bill of materials. She compares it to the lead-up to GDPR and says the reaction will be interesting to watch, but organizations need broader supply chain awareness regardless.
“In December 2027, Phase 2 goes into effect, which essentially means you need to understand your complete software bill of materials (SBOM).” Abby Kearns, CEO, ActiveState
Q: Why are open source communities split on AI-generated pull requests?
Kearns says there is no alignment across open source communities. Some refuse any AI-generated PRs, whether vulnerability identifications or fixes, while the Linux kernel now accepts them as long as they are thoughtful and attributed. Meanwhile more vulnerabilities are being identified than communities can address.
“More vulnerabilities are being identified than ever before, and the growing number of identified vulnerabilities is making it increasingly difficult for communities to address and fix them.” Abby Kearns, CEO, ActiveState
Q: Is it better to start with secure packages or patch vulnerabilities later?
Kearns says the principle is to start secure and stay secure. Scanning, endpoint detection and other defense-in-depth practices should continue, but a secure foundation avoids rebuilding packages and dependencies right before shipping. Scanning also no longer identifies all new malicious packages, so relying on it alone is a big bet.
“It’s much easier to start with security in mind, particularly when it comes to the tools you’re using to write all your software.” Abby Kearns, CEO, ActiveState
Q: What does ActiveState do to give developers secure open source packages?
Kearns says ActiveState wants developers, or AI code generators, to pull the most secure packages possible when they write code. The company focuses on the languages and on a secure place to pull from, so teams can keep their velocity and ship with confidence.
“What we’re trying to give developers, regardless of how they choose to write their code, is the best possible starting point to ensure that it’s as secure as possible.” Abby Kearns, CEO, ActiveState
Q: Should you patch open source packages yourself or wait for upstream?
Kearns says that for most modern packages ActiveState defers to upstream. It pulls down packages, rebuilds them in a secure environment, scans them, rebuilds them again in a hermetically sealed environment, scans them once more, and deploys them to a repository of choice, with daily scans for new CVEs and upstream fixes. Getting ahead of the community means backporting patches and managing LTS and STS support, which she prefers to avoid.
“I do believe that the community, particularly its contributors and maintainers, always knows what’s best and how a language should evolve.” Abby Kearns, CEO, ActiveState
Q: How can vendors help open source maintainers with vulnerability fixes?
Kearns says ActiveState identifies vulnerabilities and contributes back upstream where it can, but often waits for the upstream community to have a release available. Most identification keys off the known CVEs from CISA, so communities are usually already aware of them. ActiveState follows the community’s cadence.
“We defer to what the community’s cadence is.” Abby Kearns, CEO, ActiveState
Q: Can vetted open source components make code secure before it reaches production?
Kearns says that is the goal, but there is no such thing as 100% secure code. Vulnerabilities are found every day in everything from new releases to the Linux kernel to BSD, and much of what runs the internet is old open source code. Instead of chasing perfect security, teams should build confidence in their process and in how quickly they can update when known issues appear.
“How do we build confidence in the process and our ability to quickly respond when there’s known issues?” Abby Kearns, CEO, ActiveState
Q: What should companies change to improve AI security readiness?
Kearns says organizations do not take seriously enough the depth and breadth of the open source they use. Teams should ask which packages they pull, who is responsible for them, where they come from, and whether they are still maintained, instead of relying on a package nobody has touched in 15 years. She adds that there is no silver bullet, so teams need a collection of solutions and services.
“Open source is 98% of all of software that’s being developed today.” Abby Kearns, CEO, ActiveState
Resources & Documentation
- Curated Catalog, ActiveState’s catalog of curated open source packages
- ActiveState Library, ActiveState’s library of open source language packages
- Secure Container Image Catalog, ActiveState’s catalog of secure container images
***
👇 Click to Read Full Raw Transcript
Swapnil Bhartiya: As we all know that AI coding assistants are adding packages to codebases faster than security teams can actually patch them. So it’s, we are entering a world where there are a lot of imbalance. It’s a good thing on both sides. More code has been written. Sometimes open source maintainers complain that, hey, there is nobody to write the code now, now getting overrun. But it is also finding all the vulnerabilities that we could not find humanely. So this is a very, very timely discussion. And to discuss that, we have with us my favorite, Abby Kearns, now CEO of ActiveState. But I always used to call her the Queen of the Cloud.
Abby Kearns: We could still go with Queen of the Cloud, although, you know, maybe we could just Scratch out cloud and replace it with AI.
Swapnil Bhartiya: Yeah, without AI. I would love to just start from the very basic, is that talk about, of course, the new chapter in your career, ActiveState, the company, and what you are, of course, you’re CISO, you’re leading the organization, but what does the company do? And looking at your experience, what are you doing at ActiveState?
Abby Kearns: I could not have been more excited to join ActiveState. Particularly at this point in time, like coming back to open source, which is, you know, I’m deeply passionate about open source, the community, the way, you know, the way we’re doing innovation in the open. And I think the open source is more critical now than ever before. And with ActiveState’s desire to really give people that are writing software, either from developers or AI code generators, to be able to make sure that when they write code, they’re pulling in the most secure packages they possibly can. It just felt like a great both alignment with my history and my career, but also where I thought the moment was going, which is to say we’re writing more code than ever before, more lines of code collectively. We’re all developers now, right? I’m a developer now. You’re a developer now. We’re all writing code. However, when we write code, we’re using a language. And those languages, be it Python or Go or Rust, or even Java or .NET have packages, dependencies, and transitive dependencies. And those are things we don’t talk about often enough, but those are now aspects that are being exploited at a much faster rate. We’re hearing every week about a new exploitation that’s going through npm or PyPI. And so what we’re trying to give is developers, however they choose to write their code, the best, starting point to make sure that it’s secure as possible. But to your point, we’re at this weird, crazy moment in time, which is we’ve got more lines of code written than ever before, more open source lines of code written than ever before, coupled with the fact that we actually don’t even have alignment across the open source communities and across all the different open source projects on how we wanna even think about or deal with AI-generated PRs. Some communities are absolutely not, that we will not take or accept any AI-generated PRs, either vulnerability identifications or fixes. You’ve got some that are now leaning into it and say, absolutely, it’s fine. The Linux kernel being the most recent and the biggest that says, you know what, AI-generated PRs are fine as long as it’s thoughtful and attributed to you. And so I think we’re at this weird, intersection of more vulnerabilities are being identified than ever before, a growing number of identified vulnerabilities the communities are struggling to address and fix. And we’re at a crossroads to say, how do we think about AI-generated code in general? And so I do think this will sort itself out collectively, but right now we’re at a bit of an impasse where we’ve got the conflicts on both sides and we’re not really sure how to move forward. And every day more vulnerabilities are identified. Than the day before, just like across every open source project.
Swapnil Bhartiya: Let’s talk about, of course, depending on how people are, like Linus also, he’s now totally okay with AI. I remember a couple of years ago he was bashing it. No. No, he was like, let’s not talk about AI. Okay, let’s talk about everything else. But now he was like, and you’re right, I am now VoP code a lot, and I feel that AI has democratized a lot of things, things that we did not have access to. It is, it is. I mean, AI has been around for ages, you know, LLMs, you know, the knowledge base that they— Now, if people just flip the script, now it has democratized because now you have access to the knowledge base that you did not have access to in the past. It’s at your fingertips, at your phone, at your door. So that is incredible power now. I mean, if you look at social media, that was supposed to liberate, but now it has become addiction. People are on a reel. So every technology gets you know, abused and exploited. And so the AI is no different. What I’m trying to understand is that what does it mean for the whole software supply chain SBOM? Because for a use, for a project, oh, I will not take that code base if it is AI written. That’s a totally different thing versus that the code is shipping in production. Sure. So how do you actually track the whole supply chain and what ActiveState is doing that? Because I’m not talking about the individual developer’s problem. I’m looking at the industry-wide problem.
Abby Kearns: Yeah, I think it’s a huge problem. Like, look, I will say that, you know, we’ve been talking about software supply chain and particularly software bill of materials for 3 or 4 years now. You know, I will say that I went to an open source conference a few years ago and I was talking with a friend about that and I was like, it was when the Biden mandate came out around. And I was like, okay, this is the time. This is where people are gonna start to say, Oh, we’re using all this open source, we need to really understand that. And crickets. Like, I did not, the conversation didn’t really manifest the way that I thought. And I think to your point, we sort of swept software bill of materials and software supply chain under the rug a little bit. We’re like, yeah, it’s sort of important, we need to really understand that. And crickets. Like, I did not, the conversation didn’t really manifest the way that I thought. And I think to your point, we sort of swept software bill of materials and software supply chain under the rug a little bit. We’re like, yeah, it’s sort of important, but we’re gonna deal with that later.
Swapnil Bhartiya: But look at CRA. Now you cannot skip that in Europe.
Abby Kearns: Well, CRA, well, we’ll talk about CRA in a second. Like, but I think we, We sort of downplayed it. We’re like, okay, it’s important, we’ll deal with it later. And packages are being pulled in, and I mentioned dependencies and transitive dependencies, and transitive dependencies are something we don’t talk about very often. If you’re writing with Python and all of a sudden you’re saying, I’m gonna use NumPy or scikit, that’s pulling in packages and dependencies and transitive dependencies the developers often aren’t aware of. But they sort of were sort of paying attention. But now with AI, and AI coding assistants, they’re pulling in packages from everywhere because, and if you’ve written code, you can watch it. Oh, well, you need this Python package, you need this, and it just pulls ’em all in as it needs to to solve the problem. And so now you have even more packages and dependencies coming into your software that you’re writing of everything. And if you’ve written code, you can watch it. Oh, well, you need this Python package, you need this, and it just pulls ’em all in as it needs to to solve the problem. And so now you have even more packages and dependencies coming into your software that you’re writing of everything. And it’s coming into your core language, which means it’s part of everything. Mm-hmm. So you have more of this coming in ever before. It’s not being tracked. It’s all being shipped into production. And I think organizations are gonna have a hard time really understanding what is the breadth of my software supply chain? What is the provenance of some of these packages that we’re pulling in? And I think CRA, I think will be interesting. Obviously CRA phase 1 went into effect this past September, which is really just setting a timer on notification. Of exploitation. And so great, you’ve got that out there. In December of ’27, Phase 2 goes into effect, which essentially is saying you need to understand your full software bill of materials. So we’ll see if that is an interesting forcing factor or not. I mean, I’ve been here through a lot of these different big compliance I remember when GDPR, the lead-up to GDPR going live, everyone’s like, oh my God, we have to get all of this stuff together, we’re gonna be fined. And it was fine. But I think, like, I think I’ll be really interested to see what people do and how they react to phase 2 of EU CRA. But the fact of the matter is organizations have to get more aware of the entirety of their software supply chain. We’re gonna be fined. And it was fine. But I think, like, I think I’ll be really interested to see what people do and how they react to phase 2 of EU CRA. But the fact of the matter is organizations have to get more aware of the entirety of their software supply chain from just a pure security standpoint. More and more malicious packages are entering the stream of software development more quickly, either through malicious GitHub repos or exploitation of the package managers like npm, or malicious packages and dependencies altogether. And so you just are gonna have to build a much finer awareness on what’s coming into your environment. And one of the things that we focus on at ActiveState is the languages. And really making sure you’ve got a secure place for your developers to pull from and making sure that they’re not pulling packages directly from the package managers. A
Swapnil Bhartiya: lot of organizations who consume the package, you know, they don’t want to wait for the upstream to patch, you know, so they work with a lot of vendors, they offer, you know, patch packages, but it may or may not reach to the upstream packages. So do you work with upstream maintainers? Yes. What is your role there? That’s what I’m trying to understand. that no, the patch has to be upstream, then it’ll pull down, or no, it’s critical, let’s patch it so that companies who are relying on it, they start using it, then we’ll work on the upstream later on. How do you work?
Abby Kearns: For most of the work we do, particularly for the more modern packages, we definitely defer from the upstream. So what we do is we basically pull down the packages, we rebuild them in a secure environment, we scan them, rebuild them in a hermetically sealed environment, we scan them again, and then we like deploy them into a repository of choice. We have a catalog we can give, and you can expose that through your repo of choice. And we just maintain that. So we do a daily scan to see if there’s any new CVEs or any new fixes in the upstream. But I prefer to wait for the upstream rather than getting ahead of the community, because at the end of the day, once you— Go down that path and all of a sudden you’re applying fixes that you’re not contributing back to the upstream or you’re not waiting for the community, because I do believe that community, particularly the community of contributors and maintainers, always know what’s going to be best and how a language should evolve.
Swapnil Bhartiya: Like, you know, it’s like I would rather just really take off the kiosk of the community. Yeah, essentially it’s your distro. It’s your, like—
Abby Kearns: and I’m like, I would prefer not to do that.
Swapnil Bhartiya: Because then you’re taking a lot of debt, you know, that you have to manage and maintain. And then it’s a bit—
Abby Kearns: And then you’re into backporting the patches and figuring out LTS and STS support. Like, you know, it’s like I would rather just really take off the kiosk of the community.
Swapnil Bhartiya: So you’re the exec? Right.
Abby Kearns: And so what I’m trying to do is really defer to the community and ensure that We are supporting those communities as best we can too.
Swapnil Bhartiya: In most cases, what happens is that, of course, the codebase, I mean, it’s in production, right? So now they’re tracking vulnerabilities and it’s like, you know, reverse, you know, going back. What is the right approach? Or, I mean, it’s too early to talk about that, where, you know, that the codebase that you’re using, they’re using, is secure, you know, by default versus, you know, patching vulnerabilities, finding vulnerability. What would be the, that approach, and what is ActiveState doing to actually reach there?
Abby Kearns: Well, at ActiveState, we believe you should start secure and stay secure. So we’re not saying that starting secure is the only thing you have to do. Obviously, you should continue to scan, you should continue to do endpoint detection, all of the defense-in-depth practices that you do today, but it is much easier to start secure, particularly with the thing you’re writing all your software with. And so if you can really make sure that starting secure is the only thing you have to do, obviously you should continue to scan, you should continue to do endpoint detection, all of the defense-in-depth practices that you do today, but it is much easier to start secure, particularly with the thing you’re writing all your software with. And so if you can really make sure that you have a secure foundation, So that you’re not identifying any of these vulnerabilities by the time, right before you ship to production. Because then if you do, you’re having to rebuild all of those packages and those dependencies. But then also the fact of the matter is scanning isn’t identifying all of these new malicious packages anymore. And so that’s a big bet to make on something that is fundamental to everything you’re developing. Because then if you do, you’re having to rebuild all of those packages and those dependencies. But then also the fact of the matter is scanning isn’t identifying all of these new malicious packages anymore. And so that’s a big bet to make on something that is fundamental to everything you’re developing. And so what we wanna do is really help organizations Start their journey securely and be able to continue the velocity and ship at the pace that they want to and have confidence in what they’re delivering.
Swapnil Bhartiya: And if you look at the open source project side, maintainer side, how they can work, because right now we are talking about the downstream, the people who are consuming the project. How do you work with those? You did talk about that, but where it’s less about, hey, oh, we found the vulnerability, now we have to patch it, fix it, versus how you’re making life easy for the maintainers and the writers of the open source codebase itself.
Abby Kearns: Well, right now we are identifying those vulnerabilities, and where we can, we do contribute back to the upstream. But a lot of times we will also defer to the upstream community and say it when they’ve got a release available. Okay. ‘Cause what we’re all collectively, like when we’re identifying a CVE, we’re usually keying off the known CVEs from CISA, right? So we’re all sort of going off to that same database to say, hey, there’s known vulnerabilities. And so the community is usually already aware of those vulnerabilities. And so we defer to what the community’s cadence is.
Swapnil Bhartiya: Can you also talk about what role should vetted provenance-based, you know, kind of open source components can play in this shift where, I mean, just the way flying is, the way it’s safe and secure, right? Right. In a way, but it is not magical. There are a lot of processes behind it. Do you see, because the pace at which AI is writing code, the pace at which it is finding vulnerabilities, the time that defenders have is shrinking, you know, kind of shrinking. They don’t have enough time to actually respond, react to that. So the best thing is that just the way when you buy a car, it has airbags, brakes, everything is in place. What I’m trying to say is that do you envision a world where these processes will be in place so that once the code goes out into production, it has already been vetted.
Abby Kearns: I mean, I think that is the goal. I would say that to your point, there’s no such thing as 100% secure code, right? Like vulnerabilities are being identified every day in everything from both new releases to the Linux kernel to BSD. Like there’s like everything that we use, everything that holds up the internet and literally everything that we do in technology, has open source in it, one. And two, a lot of it is really old packages and code. And at the end of the day, I think we have to be realistic about what we’re working with. I think instead of saying like we’re gonna ever get to the point where we’re 100% confident, it’s like, how do we build confidence in the process and our ability to quickly respond when there’s known issues? And I think that is the bigger thing. Which is say, how quickly can we respond? How can we do so in a way that makes sure that whatever we’re shipping is secure as possible and we’re also able to update it quickly when we have known identifications of known vulnerabilities?
Swapnil Bhartiya: If you look at the overall picture of the discussion that we have with you, what changes companies need to make internally in their processes, in their culture? Because we always talk about culture, people, they will always, always to improve their, you know, kind of AI security readiness?
Abby Kearns: I think they need to be realistic about what does it mean to write software in today’s world. And I think I view AI as one of the tools that we have. And I think you have to really appreciate that. Well, I think one, I don’t think organizations take seriously enough the level, the depth, and the breadth of open source that they use in their organizations today. I don’t think there’s a lot of meaningfully deep conversations to say, where are we pulling? What What open source packages are we actually using? Who’s responsible for those and how are we maintaining that? Where do we pull those packages from? Are they kept up to date with active communities? Or is this one package that we’re using for this older version of Java from something some guy wrote 15 years ago and hasn’t maintained since? Like, really getting a good feel for the breadth, the depth, and the landscape of your open source. And I, obviously we’re at an open source conference, so open source is top of mind. But also I really wanna reiterate, open source is 98% of all of software that’s being developed today. It’s in everything. And so if you don’t understand the bulk of what everything you’re running is, then that’s a big blind spot. And so really getting awareness of what you’ve got, where you have it, how you’re using it, how do you maintain it? Not maintaining it, I would probably start there and say, what are our best practices and the processes that we can apply? And then moving from there, how do we make sure that going forward we’re leveraging it securely, we’re leveraging it successfully, and we’re able to write code that we feel high confidence in that we’re able to maintain going forward? As an industry, we’ve been looking for the silver bullet for decades. There’s going to be that one tech that’s going to solve all of our problems. And I think it’s suffice to say there’s no such thing. And so understanding the collection of solutions and software and services and technologies you need to leverage to build the right solution for your organization and to really be thoughtful about that, but instead of looking for the one thing that’s going to solve everything.
Swapnil Bhartiya: Oh, that’s very well said, Abby. Once again, thank you so much for your time today. And as usual, after a long gap, I would love to have you back on the show and continue this discussion. Thank you.
Abby Kearns: I just count myself privileged to be part of the TFiR Network again.
Swapnil Bhartiya: Oh, the honor is all mine, you know, to have you back on the show. And AI, unlike other technologies, is going to be there for a very, very, very, very long time.
Abby Kearns: It’s not going away.
Swapnil Bhartiya: It’s not going away. I think just the way we used to say that all code will be open source, I think now AI is also another major layer which is never going to go away. So I will, I will, I will, so I see that, you know, we’ll talk a lot. But once again, thank you so much.





