CyberWire Daily - Contents may be malicious. [Research Saturday]

Episode Date: October 10, 2026

Adam Meyers, Head of Counter Adversary Operations at CrowdStrike, is discussing their work on "PhantomRaven, An LLM-Generated Information Stealer Developed for Bug Bounty Hunting." PhantomRaven, a JavaScript-based information stealer distributed through malicious npm packages by a financially motivated threat actor posing as a bug bounty hunter. The malware targets system information and continuous integration and continuous deployment (CI/CD) environment variables, likely seeking credentials, with analysis suggesting its code was generated using a large language model. The report explores how AI-generated tools may lower the barrier to cybercrime and outlines steps organizations can take to mitigate risks, including restricting package installation scripts, using private npm registries, and monitoring dependencies. The research and executive brief can be found here: PhantomRaven: An LLM-Generated Information Stealer Developed for Bug Bounty Hunting Learn more about your ad choices. Visit megaphone.fm/adchoices

You're listening to the Cyberwire Network, powered by N2K. And now a word from our sponsor, Trinity Cyber. Detection and response was failing before AI. Now it's breaking under the speed and scale of AI-driven threats. That's the reality Trinity Cyber was built for. The Trinity Cyber platform analyzes network sessions in real time, removes malicious content in line, and provides unified control across,

users, applications, devices, networks, and AI. By bringing protection and policy together, Trinity Cyber helps security teams reduce risk, simplify operations, and stop threats before they become incidents. Learn how at trinitycyber.com. Hello everyone and welcome to the Cyberwires Research Saturday. I'm Dave Bittner, and this is our weekly conversation with researchers and analysts tracking down the threats and vulnerabilities, solving some of the hard problems and protecting ourselves in a rapidly evolving cyberspace. Thanks for joining us. We had seen the tool that this threat actor has been deploying was kind of the real impetus for us to start tracking it. So the tool that they've been using for a while now has been what we

would call an info stealer. So that kind of started us down the path of research. this threat actor. And what we started to see was that it was to pooling that was malicious. And that's kind of how we picked up the Phantom Raven information stealer. That's Adam Myers, head of counter-adversary operations at CrowdStrike. The research we're discussing today is titled Phantom Raven, an LLM-generated information stealer developed for bug bounty hunting. Well, tell us about the person that you suspect is behind Phantom Raven.

even, you feel like this is probably a bug bounty hunter? I think that would be a very generous way of characterizing them. I think that they're taking advantage of bug bounties using this info stealer so that they can get paid. So they're monetizing, they're using a tool that it's malware, it's malicious, and they're using that to then allow them to monetize bug bounties. and that's kind of the, I wouldn't say that they're necessarily legitimate bug bounty hunter as much as they're kind of gaming the system.

According to the research, the threat actor reportedly claimed to have found vulnerabilities by publishing malicious packages and getting code to execute on other people's systems. Can you contrast that for us, compare that to legitimate bug bounty research? Well, I think it's important, to note that there's different types of bug bounties out there. So in some cases, you know,

it's a technical bug bounty where you're looking for a flaw in the software. In other cases, it is looking for, you know, some way to gain access to information. So, you know, comparing it to a typical bug bounty hunter, they're going to take a look at software, try to find vulnerabilities and then go through the bug bounty program as a way to conduct responsible disclosure. What this threat actors doing is they are distributing this JavaScript tool that we call Phantom Raven via Node Package Manager. Then once they gain access to somebody through this MPM, they're taking that information and using that as part of the bug bounty. They're contacting organizations, claiming to have identified as a compromised device, and then use that to get paid out on bug bounty.

Can you take us through the Phantom Raven infection from the developers' perspective? What do they think they're installing and then what happens behind the scenes? Yeah, so this gets to, you know, a lot of the supply chain attacks that we've been seeing over the last year is organizations have heavily become dependent on NPM for Node. So Node.js is a very popular development, language tool kind of ecosystem, I guess would be a better way to think about it. And they're gaining access to these developers by, you know, types of. of squatting a package, right? So if you expect that a package is called Transform dash JSBI, dash 2, big int, or something like that, and you mistype that, then you're going to end up with this package, and you've now deployed the malicious package accidentally.

So you note in the research that the NPM packages themselves don't have a very much malicious code in them, and they're using a remote dynamic dependency. Can you describe how that works for us? NPM profiles are, they contain these files that include these remote dynamic dependencies, and they tie back to a domain that is, we believe, the command of control domain for this threat actor. So that's kind of the way that they deploy this Moist JavaScript that gets executed, and starts running, and then it calls back to that domain via the RDD.

And what information does it collect once it's up and running? It will collect system information. So what operating system, host name, IP addresses, it'll get process and various runtime information, the directory it's running from, process IDs, what version of Node.js they're running, various command line arguments, username, date and time, and then all of the build environment variables.

So project path, the version, etc., GitHub actions, things like that. So quite a bit of information gets collected by Phantom Raven. You know, one of the interesting conclusions in your research was that this was almost certainly generated with an LLM. What was the evidence in the code that led you all to that assessment? We've seen this evolve over the last couple of years where threat actors who have started incorporating LLMs into their workflow, the stuff that they're using, the code that they develop ends up having much better comments in it. They have better error handling. We've looked at, you know,

one of the things that we do is we do statistical token analysis patterns across these scripts to figure out what is the likelihood that it was generated from an LLM. And, you know, so there's a couple of things like if you take a look at the code, you could see that it's got a comment before every global variable, for example, in the code, which is something that threat actors don't need to do. In fact, it's kind of extraneous, but LLMs do that so that user that asked it to generate the code will have a better understanding of what they're looking at. So before every single global variable, you'll see collect CICD and environment information or collect system information or fetch public IP dynamically. So as it's, you know, writing the code for the threat actor,

it's making better code than the threat actor themselves would likely have done. So being overly verbose with the comments. Exactly. We'll be right back. DLP sucks. Every CISO knows it. 20 years of rules and reg X built for another era. Jazz is the DLP that deeply understands your business.

Melody, Jazz's agentic investigator, weighs every data, movement against the data, systems, people in process, then provides analysts with the few incidents fully investigated. In 30 days, 2 million signals in, about 80 investigated incidents out. No rules written. Jazz won the 2026 Crowdstrike AWS and NVIDIA startup accelerator from a nearly thousand applicants. See Melody in action at jazz.security slash N2K. Yes, you can have an enterprise network that's secure and reliable and high performance. And no, you don't need to choose the best two out of three. With Meter, you can get the end-to-end network built from the ground up,

fast to deploy, and easy to manage. That's because Meter is software-led for easy installation, maintenance and control for everything running on your enterprise network. Hardware, firmware, and software all working together from the start seamlessly on a unified platform that's secure by design. You can't protect what you don't know exists, which is why meter gives you comprehensive visibility into wired and wireless routing, switching, firewalls, DNS security, and VPNs.

You'll really know what's running on your network down to the most granular client level. Step off the hardware box upgrade treadmill and switch to meter for a predictable fee and free up your team to spend time, on all the other things that keep your business running. Try it out for yourself and book a demo online at meter.com slash cyberwire. That's M-E-T-E-R dot com slash cyberwire. Now, you all assess the author's technical sophistication is relatively low. I'm sure that must sting for them. But how do you

reconcile that assessment with the fact that despite that lack of technical, sophistication, they did build a functioning information stealer here. Well, I mean, they told the LLM to build this functioning information steeler. Fair enough. Fair enough. And look, it doesn't matter if they're sophisticated or not sophisticated. What matters is, did it work? And are they monetizing it?

And I think that that's probably the measurement of success here and not our assessment as to their technical sophistication, though hopefully it does sting a little bit. But, you know, I think that that's, you know, one of the important things to keep sight of is, did it work? And if it worked, then it was good enough. Does the quality of the malware actually matter here? If the delivery mechanism works and the stealer gets access to the secrets, I guess is it fair to say the code doesn't have to be particularly sophisticated? It doesn't have to be sophisticated, but it does need to work and be reliable for the threat actor to be able to use it. So, you know, it's one thing if they write code and it's poorly documented and a little bit hacky,

which in this case, you know, they used an LOM so that wasn't the case. But if they had and, you know, the data got corrupted when it was being stolen, then it's no good to them. So it matters in as much as the outcome, again, is that they get access to the information that they were looking for. What are your recommendations to developers here in terms of defensive measures? Well, I think this is a much bigger issue that people need to be thinking about in terms of CICD. And I think it's relatively poorly understood for most folks that they have these continuous integration, continuous deployment environments where if somebody pushes code into a Git repository on the other side of the world, right, this is kind of like that, that whole chaos theory or something like that, a butterfly flaps its wings, right? Somebody pushes code and then it gets automatically consumed and deployed. through your CICD environment without you knowing that that happened, you've now deployed

most code into your environment, and that's obviously not good news for you. So one of the things that we recommend is implementing a cool-off period, meaning that new packages can't be deployed until three days after they were first published. So it gives you a little bit of time to put between when they, that code was published and when you start using it. The other thing, you know, that's important to be careful of is that you don't typo-squat these node packages and or fall victim to typo-squotted node packages, I should say, where, you know, if you fat finger or something, you end up pulling down the wrong package. And this isn't just node, I should note it. You know, we've seen this with

Python and we've seen this with other repository ecosystems. And that is something that has been going on for a while. So making sure that, you know, you have constant checks to look for the approved node packages or the approved Python packages or whatever it might be will ensure that you can detect when these potentially malicious

packages were errantly deployed. Does the fact that you're seeing more and more of these AI generated malware packages, does that change your approach? Should it change what defenders are out there looking for? No. I mean, this is, this is

been going on now for the better part of two years. We've seen more and more threat actors, you know, just like everything else in life, we've seen AI kind of popping up in all of these places. So threat actors have been using LLMs more and more to build tools, to build scripts, to build malicious code. You know, I think that the important thing is not necessarily if the code is written better than a human would have written it. It's still doing the same thing. So you need to have the right visibility into your environment to be able to detect when that code gets instantiated and what it's doing. What do you hope the takehomes are for folks who check out this research?

What do you want them to walk away with? Well, a couple of things. I think, you know, this is a pretty novel approach for how a threat actor might abuse the various bug bounty programs that organizations implement. You know, they go out and they do these things for trying to make their, their platform, their systems more secure, you know, it's important for them to understand that, hey, like, threat actors might be abusing that and you need to be careful. You know, we've seen other systems that have been targeted as a result of some of these

these bug bounties. You know, there's been other kind of things that we've picked up, other adversaries that maybe we'll cover in the future. But it's important, I think, to understand that the capability exists. And more broadly, I think it's important to really kind of shed weight on the fact that you're pulling down this code and particularly in the CETA. CICD environments, you need to be very careful about how you implement that, putting policies in place, and having adequate checks to make sure that you're not falling victim because these types of things can get into an environment. And if you don't notice it, they could be there for months or even years before you realize that you have a problem.

It's a really interesting insight. It seems like there's this pressure to move faster, faster. but there are some things like this where pausing for a little while maybe the best approach. In terms of rolling out the packages. Yeah. Yeah. I think that we've seen these supply chain attacks. We mentioned this in our threat hunting report just a few weeks ago that in one case we saw a clip that we call altered spider, which a lot of people know is team PCP,

that they were able to, in fact, over 300 repositories. in a single day. So, you know, this is something that has been an increasing problem. We've seen a number of these supply chain attacks conducted by threat actors from North Korea, criminal threat actors, and other groups targeting these supply chains. So it's a dangerous space out there, and you need to make sure that you have the right controls in place and that you're thinking about what the risk model looks like by using some of these environments and some of these repositories.

Our thanks to Adam Myers from Crowdstrike for joining us. The research is titled Phantom Raven, an LLM-generated information stealer developed for bug bounty hunting. We'll have a link in the show notes. And that's Research Saturday, brought to you by N2K Cyberwire. We'd love to know what you think of this podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like our show, please share a really. rating and review in your favorite podcast app.

Please also fill out the survey in the show notes or send an email to Cyberwire at N2K.com. This episode was produced by Liz Stokes. We're mixed by Elliot Peltzman and Trey Hester. Our executive producer is Jennifer Ivan. Peter Kilpe is our publisher, and I'm Dave Bittner. Thanks for listening. We'll see you back here next time.

And now a word from our sponsor, SpectorOps. Today, AI is rapidly adding non-human and agentic identities to modern enterprise environments, creating new trust relationships and attack paths. Bloodhound Enterprise helps defenders map attack paths across AWS and hybrid environments as one connected graph, identify the choke points that matter most, and bring trusted attack path intelligence

into approved AI workflows with Bloodhound Hunter. See how SpectorOps helps teams secure the AI-driven identity era at specterops.io.