Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

The OpenSourceMalware Show

The OpenSourceMalware Show artwork

Trojanized Claude Code extension, Binance and Polymarket thefts, PolinRider dated to 2025

Nov 6, 202647 minAudio
ShareXLinkedInFacebookWhatsAppEmail

This week we talked about: Trojanized Claude Code extension: On October 29, a trojanized Claude Code extension was shipped from Cursor's own mirror of the Open VSX registry. Cursor's mirror can carry different versions than Open VSX itself, so the only people who got it pulled Claude Code through Cursor, and most of them appear to have had auto-update turned on. The extension included the official binary plus code that gives attackers persistent remote access. Binance robbed by North Korea: The $6.3 billion Binance crypto theft started in June, when an engineer downloaded a malicious version of Homebrew. Binance cleaned it up, but the attackers had already learned enough about that engineer and his colleagues to pick their next target. In September a second engineer was compromised through a fake job interview's take-home assignment on a work laptop.. That access led to two more developer machines, and a container deployment token on one of them was used to deploy the container that launched the attack on October 29. Polymarket compromised through DebugBear: Polymarket was compromised through malicious JavaScript from DebugBear, a third-party vendor whose DNS records were hijacked at its domain registrar. Beyond DebugBear's first incident post on X, there's no public information about the attack yet. Injected JavaScript isn't new: it has been a common WordPress incident for years, through sites with an upload function and path traversal. PolinRider backdated to 2025: A set of legitimate npm packages compromised in May 2025 carried an obfuscated payload that turns out to be the earliest known PolinRider infection. The same maintainer was hit again multiple times this year, indicating they were likely compromised the entire time and could be one of the earliest victims. Resources (blog) Trojanized Claude Code Extension Plants an SSH Backdoor (blog) Security notice: malicious version of Anthropic.claude-code (2.1.312 and 2.1.313) identified in the Cursor VS Code Extension Marketplace (blog) 29 October 2026 Binance Security Incident: Initial Findings and Response Update (webpage) North Korea Responsible for $6 Billion+ Binance Attack (blog) SEAL Releases Advisory on DPRK Threat to Crypto Exchanges (article) Polymarket Returns $3.9 Million to Traders as DebugBear Cites DNS Hijack (blog) Earliest Known PolinRider Malware Found in 2025 npm Packages

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Jenn GileHello, it is Thursday, November 5th. I'm Jen Gil and I have Paul McCartney with me. We are the co-founders of Open Source Malware. And this is the Open Source Malware show where we talk about the week's malicious open source news. And wow, Paul, do we have a lot to cover today? Just in brief, we're going to talk about the Cloud Code extension. We're going to talk about Binance. We're going to pop over to Polymarket for a little bit and we're going to talk about Paul and Writer. And as we were discussing, my show notes today are like, they're like a novel.

Paul McCartyThey're like three times long, four times longer than they normally are, right?

Jenn GileYeah, usually they're pretty sketchy. And uh, you know, we make it, but a lot has happened in the last week. And so let's start with last Thursday. I was down in Austin. I was kind of getting ready for bed when I get a message from you that says something along the lines of hold on to your pants, something big is coming. And uh to kind of you know give everyone an idea of what we were going through last Thursday, my Thursday or Friday, on October 29th, a community member tipped you off about a trojanized VSIX package.

Jenn GileThat's the standard format for VS Code and OpenVSX. So it was version 2.1.312 of the official Claude Code VS Code extension. It contained the official binary and also a little treat. Uh, it would give attackers persistent remote access to the machine where it was installed. It didn't actually um have an info stealer in it, but it was looking for cloud credentials. And when we first saw this, we knew it could be really, really bad, but we had like not a lot of information.

Jenn GileAnd what made this complicated is there's a like um it's not always clear based on the format where something might have come from and how far it might have spread. And so we were really trying to figure out in those early hours which vendor had been compromised. You know, was it Claude code and anthropic? Was it one of the marketplaces? Like, where did this happen? Was it a third vendor? Um, what we saw was that the official Claude code packages for the VS Code marketplace, the Open VSX marketplace, and NPM, all of those skipped versions straight from 2.1.311 to 2.1.313.

Jenn GileAnd there were no statements from any of those uh platforms. Correction, including, oh sorry, jump in.

Paul McCartyCorrection. So yeah, sorry. They skipped from 3.1.2, which was the last legitimate version. They skipped to 3.1.4. 3.1.3 is the one that was missing.

Jenn GileApologize. Uh apologies. Yeah. So uh there was a version skip. What we were doing, both of us, we were looking at all these marketplaces and trying to see was there malware alive somewhere? And all we saw was a skipped version. There were no statements from any of the vendors, including Anthropic. And so we kind of had to start from the place that a lot of you know incident responders would start from of assuming compromise across all of these. And so what we did was we sent an alert to our customers and we posted on social media kind of a quick, hey, we don't really know a lot right now, except there's a compromised official version out there.

Jenn GileIt could have gone out through any of these channels, immediately look for this version while we try to figure out what's going on. And so uh I'll pause there. Like you were in the meat of this. Tell me what was going on.

Paul McCartyYeah, I was, I mean, it doesn't happen much more real time than how it happened that day. So I mean, I think the really important thing to start with is that we had the artifacts in our hands, right? So, like we got from the victim, we got the 3.1.3 um version of you know, what looked like an open VSX um uh extension. So we had it. Um and we knew that um the you know, for all intents and purposes, it was the same thing as 3.12 because we it was easy for us to compare it against the the earlier version and to see that it had this um you know additional file that the you know where the the payload was.

Paul McCartyUm so I think that's the important thing to start with is that we had it. So we were looking at an open VSX, I mean, it was an open VSX uh package at the time. And you know, I I'll jump right in here. I think at the time, um, you know, like you said, there was no information, and this happens all the time. When we find something new, um, you know, you're out there on your own, kind of trying to figure this out and do research, and you're dealing with the victim and you're getting victim accounts. Um, and that's the other thing, too, is that the victim was saying it was open, you know, they got it from open vs.

Paul McCartySo you know at that point it was a it was a done deal. It was it was an open vs.

Jenn GileAnd it was, it just wasn't from the main open vs um well yeah, but we were able to eliminate some sources, so we determined it was not pushed through the VS Code marketplace, we determined it was not pushed through the npm marketplace, and so based on the information you had, we determined okay, it looks like it came from the open VSX marketplace. That's what we're gonna share out with people, and then things got even more interesting.

Jenn GileUm, what can you share about the back channels? I don't want to uh say anything that uh will get me in trouble.

Paul McCartyYeah, well, I want to do the same thing, so I'm gonna be very um uh I'm gonna I'm gonna choose my words very carefully. Yeah, so basically at that point, we were the only blog out there um talking about this. We were the only researchers that knew about this. You know, all the other big companies worth a billion dollars had no clue. Um so uh what happened is a friend reached out um and and act, and this happens all the time. Like I do disclosure constantly, and this happens all the time. I will often be reaching out to somebody that I know is an intermediary.

Paul McCartyIn this case, it went the other the other direction. Somebody reached out to me, uh, put me in contact with the cursor team, and one of the founders of Cursor uh gave me a call, and we had like a little mini um, you know, kind of incident response kind of um uh you know chat about what was going on. Basically, there and none of this is like, you know, I haven't signed any NDA, and and you know, I mean we can talk about this pretty openly, but the basically what they asked me was you know, where have you seen this?

Paul McCartyAnd I said, well, at this point, all I have is this open vs extension. We and we were pretty sure that this never hit npm um or the mic the Microsoft VS Code marketplace either. Um and uh cursor said okay, so you've only seen an OpenVSX. And yeah, like did you do you were you able to download the package from OpenVSX? And I said, No, I I was not. Um now the problem is we had all these kind of artifacts coming in from different you know victims, and you know, we're trying you and I are trying to manage all this in real time, uh, you know, when we're on opposite sides of the world from each other, which is always fun.

Paul McCartyBut um they basically uh cursor uh then said, Hey, we think this is actually from us, and then they started talking about theirs. Now, up until this point, I didn't realize that cursor had a separate open vs mirror. You know, some other people like Raphael, for example, at Akido, um, who you had lunch with recently, right? Yeah, great guy. Um uh knew that. So other people understood that kind of relationship better than I did. But yeah, Cursor then um explained that um they had their own OpenVSX mirror, and then and they also alluded to the fact that it did some interesting things, right?

Paul McCartyAnd in the meantime, you know, I have agents working in the background, and I'm getting stuff coming in from some of our other friends, and you know, it's all everything's all really, really fluid. But I think the important thing to note here is that cursor was looking from me to identify whether or not I had actually found it in other places. So I think they didn't, they weren't 100% sure it was inside of their organization yet. And I don't think they understood how it got there at that point. Um, so you know, when I said, Okay, hey, you know, everything we're hearing is open VSX, they said to me, Well, we think it's cursor.

Paul McCartyAnd I said, Well, hey, listen, I've got some of the victims on the phone right now. Let me just reach out to them and ask. And so I went through the instant responder and I said, Hey, can you ask the victims directly? They kept saying open VSX, but can you verify if they it was open VSX or for cursor? And then really quickly, I got updated details which were saying, No, this was cursor auto updates versus cursor. At least one people saw it, one person saw it happen. And you know, that's when things started to change. Our frame of reference at that point changed.

Jenn GileYeah, so like to clarify for anyone who has uh like I feel like the always sunny Philadelphia Charlie meme where he's like pointing in five different directions and like looks crazy. Um there the cursor maintains a limited mirror of the open vs registry, which contains or can contain in this case separate um versions from what's actually an open vs.

Jenn GileSo the only way you would have consumed this extension would be number one if you use cursor specifically to pull a Claude code extension. So you're not getting that extension from somewhere else, you're consuming it directly through cursor's uh ecosystem. And number two, if you had the auto update enabled, I mean, granted, somebody could have uh manually you know chosen to get the malicious version, but it wasn't up a super long time.

Jenn GileAnd so from what we understand, most of the people who were affected had auto update turned on in their cursor extension library. Yeah.

Paul McCartyYeah, correct.

Jenn GileUm and Paul, I'm gonna pause you for a second. We've got somebody on YouTube saying there's a little bit of mic feedback, so make sure you don't turn your head away.

Paul McCartyGot it. I will laser focus on my microphone.

Jenn GileI know. So anything else uh we want to talk about with regard to this uh trojanized extension.

Paul McCartyUm I think that the important thing, you were touching on it there um before we s we made that change, is that there are two um there's two uh kind of internal ways that cursor will feed out extensions to their users. One of them is that they will effectively proxy Open VSX. The other is that they have a set of internal um packages, sorry, extensions that they will feed out.

Paul McCartyAnd those internal I I enumerated um a whole bunch of them um and I found uh I think 12 that in particular that are strange, um, which you know I'm putting this all on a blog post soon. But I think the thing that's important to understand here is that inside of their system, if you ask for, hey, let me ask for gen you know extension version 1.0.0, it'll go to this internal service, which is called Gallery, and then Gallery will decide where do I get this actual extension from? Do I get it from our OpenVSX mirror or do I get it from this package repository?

Paul McCartyAnd there's logic there that'll tell them which things happen. Now there's a whole bunch of other logic there that does some weird stuff too as well. So for example, cursor was originally owned by a company called AnySphere, and there are rules inside of the cursor gallery service that says, hey, if you ask for an any fear, any sphere service, guess what? Don't check it with your your your firewall, sorry, your um your malware scanning, and there's all kinds of this other kind of thing.

Jenn GileOh, interesting. I didn't catch that one, huh?

Paul McCartyNo, yeah. I mean, I the the I haven't pushed the blog post yet to the Intel library, but um the there's just a lot of weird stuff going on there, right? And so they they built this years ago, and it was feeding out extensions, and you know, then they added the Open VSX mirror thing, and it just all got really kind of complicated behind the scenes. So another thing they do is they serve out a version of the Microsoft Python uh extension that's old, and they do that specifically because the new version, the official version, is isn't compatible with the new version of cursor.

Paul McCartySo they're they're serving out this early 2025 version of the package for if you ask for the the official Python um extension, they'll they'll hand you this old one, even though there's a new one in there. So they have this logic inside of there that says, hey, if a customer asks for the Python extension, this this Python extension, don't give them the one from OpenVSX, give them this one from our internal storage. I'll I'll cap it there. There's there's a lot more that's going on behind the scenes, but we'll leave it there.

Jenn GileNo, uh I think we continued. Uh the where we'll leave this is it's easy to uh second guess or Monday morning quarterback or whatever version of that. Um, you know, how people handle these. I think we stand by the initial statements about, you know, look for it across all of these ecosystems, because in the lack of any statements from any vendors, you know, assume compromise.

Jenn GileI I would have liked to have seen something from these ecosystems. You know, they intentionally skipped versions, and that implies a choice was made. So, you know, not great, but at the same time, I also do understand how corporate uh comms and legal requirements tend to work with these kind of statements, and they probably didn't want to put out a statement until they knew what they were saying because nobody loves an I don't know what's happening statement.

Jenn GileThe lawyers don't like it, the comms people don't like it. So I don't know. I see both sides, don't love how that went down, but I also understand it.

Paul McCartyYeah, I'm gonna double-click there. I think that anthropic was definitely involved in this because all of those packages, you know, to skip version 313 in NPM and VS Code Marketplace and the other places, that was anthropic. That was not cursor. So obviously, cursor was talking behind the scenes to anthropic, right? Neither one of them said anything until our blog post came out. And the and my assumption is that cursor would not have made anything, many made any of this public unless it was for our blog post. In fact, I've actually heard that you know explicitly said in the grapevine.

Paul McCartySo um uh I and I am disappointed by the fact that these companies continually do this kind of yeah, it's just it's it's I don't like that the AI companies in particular are not honest and transparent about these kind of security events because a lot of us are running these tools and we're you know we're downloading components inside of these ecosystems, and that fills me with a lot of distrust in these companies.

Paul McCartyAnd I'm I know a lot of and that's why this blew up because a lot of other people have that feel that same kind of distrust.

Jenn GileYep. Amen to that. Um we were talking, you know, throughout this week about what we would do again and what we would do differently. Like I said, I you know, hindsight being what it is, I would still make the same decision to flag that all of those ecosystems were potentially delivering malware. Paul, what would you do the same again? And what would you knowing what you know now, what would you do differently?

Paul McCartyYeah, I mean, I think that there were a bunch of, I'm gonna be a little spicy here. There's a bunch of arm armchair quarterbacks coming from other companies who didn't have this, right? And so um sucks to be them. But um, the the reality is that uh there are some things that I would do differently. Um I think in particular, when you're when you're in the incident response mode and you're trying to, you know, you're trying to work through this, you're getting a lot of data coming in from a lot of different sources, right? And you're trying to figure out how much of that is true and how much of that is true, and how much of that is true, and you're kind of putting this together in this kind of you know, this narrative, right?

Paul McCartyAnd it's very disjointed. I think sometimes my inclination personally, having done incident response for 30 plus years, all right. So please, like all these all these people new to the industry, please. I I've been doing this for a while, and the reality is that it's my job. People use OSM to tell them when something's malicious, and it's my job to make sure that we try to protect them as much as we can. And in that moment, the thing that I would do different is probably I would have held back a little bit longer on you know making some of those kind of judgments about um npm or otherwise.

Paul McCartyBut at that point, we thought we were you know protecting our customers, and we were, and so to to be very explicit, I would absolutely do the same thing again. If I see a missing version of npm 3.3, you know, or VS Code extension, I would do exactly what I did, which by the way, is I said, hey, listen, these versions are missing, they're presumed malicious, but we don't actually have data. We have not inspected. We always said that, so there's always these big qualifications around it. So I would absolutely 100% do all that again.

Paul McCartyIt's my job to help people make sure that they don't ingest malicious stuff into their uh into their applications and their environments, and and that was part of that.

Jenn GileYeah, yeah. Hard decisions to make, definitely no fun uh making those decisions under time pressure. Like it is what it is.

Paul McCartyAgreed.

Jenn GileOkay. Let's move on to the other huge news story. This is one that we were not at all involved in. And so uh this is something where I'm gonna uh talk about what happened and then let's kind of analyze it. So, unless you've been living under a rock, you've heard that Binance uh got hit. It is being called the biggest bank robbery in history. I didn't validate that particular claim, but it's certainly the biggest crypto theft in history.

Jenn GileAttackers stole$6.3 billion in crypto, or at least that's what they tried to steal. Roughly$3 billion of it was frozen by um tether and circle. And so they may not have been able to make off with the full amount. There's a lot happening right now to try to recover those stolen funds. Um, but immediately rumors started spreading that it was related to a compromised developer. And very quickly, the FBI linked it to North Korea because it shared infrastructure with the Bybit attack, which happened earlier this year, very similar um shape to it.

Jenn GileUm, like we've seen with other attacks, reusing infrastructure is not uncommon. Um, these are not things that are constantly firing. And if this is infrastructure that uh can't get shut down, then it doesn't, you know, there's not a lot you can do about it. So that's all the you know setup to this. On November 3rd, Binance released a um kind of a retrospective, a statement about the compromise. And I think generally they did a nice job with this.

Jenn GileThey definitely admitted that they uh failed, uh, that there were failure points. I know you would always like to see more. Um but I'm gonna talk about it. I know we are. Um, but they separated it into both how the attack itself played out as well as the kind of like building blocks that the attackers had in place. And so I'm gonna start there because um I don't know. I'm biased. That's always the part that I find most interesting. Um, so this started in June of 2026 with one of their engineers um doing a browser search for homebrew, didn't really pay attention to where they were downloading it from, downloaded a malicious version.

Jenn GileUm, it seems, according to Binance, that they detected that very quickly, cleaned everything up. However, uh it triggered for long enough that the attacker was able to get some information about what that engineer did and his colleagues. And so this, you know, whether that uh that was not a like targeted attack. This was certainly luck on the part of the threat actor that an engineer clicked a bad link.

Jenn GileHowever, what they did say is throughout the year, they have detected and prevented, they said 57 suspected North Korean linked cyber threat cases that span the gamut of all the stuff you and I have been talking about on this podcast. Fake job interviews, uh, malicious software, uh Compromised packages and browser extensions, fake video conferencing applications, social engineering campaigns like finance has been getting hit all year, which is no surprise to us, given you know they do crypto and it's a great target if you're a threat actor.

Jenn GileUh fast forward to September, they evidently used that information that they got from their uh lucky attack. What do they say? Luck favors the prepared. Uh, so they got enough information to figure out which engineers to target with the social engineering component of the scam. And so the next engineer gets hit with what we know as contagious interview. They didn't describe it that way in their incident report, but from reading through it, it's very obvious that's what it was.

Jenn GileAnd we've kind of heard some, again, uh chatter from circles that confirms that's what it was. Um, but somebody reached out to the engineer, uh, offered them a job interview. They went through multiple loops, uh, they were shown lots of things to raise their trust in the process that they were in a legitimate interview process. The trust raised enough that when they were given a take home code assignment, they went ahead and completed that.

Jenn GileNot only did they complete it, but they completed it on their work laptop, which uh goes to show it happens. It happens more than we want to admit. Um, but that take home code assignment was a repository that contained malicious code. And um we don't have a whole lot of detail on exactly what it was, but in reading through the components that they describe, it sounds very much like the classic kill chain that we've been documenting all year.

Jenn GileUh, it had a component that um replaced the Pi and V with a malicious version, which we've seen that same pattern with them replacing the NPM CLI with a malicious version. I can almost guarantee that this um person had the Paul and Writer campaign on their machine, very likely passed it around. Uh, but Binance says their endpoint detections didn't pick up any of this activity, which tracks with what we've seen.

Jenn GileEDR cannot catch this type of attack because it's not anomalous behavior, it's normal software engineering behavior. And so they didn't say specifically when this happened in September, but it sounds like the attackers had access for well over a month before they started making some moves. So on the 20th, they were able to use that compromised engineer to get access to two other developer machines.

Jenn GileOne of those developers had a container deployment permission token on their computer. And then on the 29th, so also last Thursday, last Thursday was a busy day. Um, they used that credential to um remote access deploy a container, and that container triggered a malicious script, which launched the attack. So that's just the setup. Like we didn't even talk about the attack. Um, this has been an interesting one to track, huh, Paul?

Paul McCartyIt it has. Um meanwhile, SEAL Alliance came out with um a really good article talking about um the Binance hack and the difference between um Contagious Interview and Traitor Traitor. And I think that is an important thing to highlight here because as of yet, there's no evidence that Pollen Rider was involved in um you know, in this Binance attack. But if Pollenrider was on one or some N number of Binance laptops, I that gives DPRK the ability, because they've got persistence and visibility on these laptops, the ability to do you know enumeration and figure out, oh, who is it here that we want to specifically target inside of this organization?

Paul McCartyBut not knowing if Pollen Rider is on there or not, I think it's important to talk about the the tools they use. So this Pi um ENV um uh attack component they used is not part of the typical um Pollen Rider and and Contagious Interview toolkit.

Jenn GileYeah, and I wonder if that was like a little bit of a custom thing given they had some intel on what Binance was using.

Paul McCartyI think this is one of the distinctions that SEAL is making between contagious interview and trader trader, in the sense that that you know, like whether like we we used to talk about blue nor if and now we're talking about trader trader, but in both those cases, the idea is that at some point it's passed off from you know these lower teams, the B team to the A team, and then the A team has the skill on their team to be able to customize some of the stuff, or at least pull out tools that you know they're not just throwing at everything like they are with the common kind of pollen writer stuff.

Paul McCartyLike we, you know, we we've mapped dozens and dozens and dozens of campaign IDs um and the fingerprints for each one of those for the pollen writer stuff. Um, so we'll pick up on that quickly. So it makes sense that they would use, even if it's not genuinely custom for this one-off, at least a toolkit that you don't see that they only pull out for specific, um, kind of more you know, one-on-one customized attacks.

Jenn GileYeah. Well, I mean, the takeaway here uh that I think you left in the show notes is that one developer led to the largest crypto heist in history. Um, my takeaway is it has been challenging to help people understand how malicious open source is just as, if not more threatening than your classic ransomware, because people really pay attention to ransomware. They think about that dollar figure. So I guess my hope is that people learn from this and can see the real dollar outcome that can happen and can perhaps then take a little bit of a leap and say, okay, we don't we don't have crypto on our platform, but we do have things that are worth something that this kind of an attack could make possible to exfiltrate.

Paul McCartyYeah, one thing I just want to double click too is you you briefly highlighted it, but I think um Binance's timeline is good in some places and pretty crap in other places. Um, for example, and I know other people have talked about like for example on the Risky Biz podcast, they were talking about this, but I agree with them. You know, after it gets to that um that component where they get access to the the first developer's laptop via contagious interview slash trader trader, then it gets kind of muddy about what happens next.

Paul McCartyAnd they don't go to into any detail whatsoever. And I think one of the things that's that that means is that when people are talking about this attack, people are saying, oh, it's very simple, you know, relative to the the BitGet attack, this is relatively simple. And I think some of the complexity that has probably happened in this attack is getting lost because Binance hasn't gone into detail about it, right? So right now, it just I've hey Binance, by the way, what's happening is the narrative out there is that you had a developer that got popped during contagious interview and then six billion dollars left your organization.

Paul McCartySo I think it behooves you to go into a little bit more detail about what happened after that initial contagious interview/slash trader trader compromise because that'll help us understand the narrative better. Because right now people are oversimplifying it, and I don't think that's good for anybody. It's not good for Binance, and it's not good for us, the ecosystem.

Jenn GileYeah, well said. Okay. In brief, I think it's worth talking about polymarket because it happened basically the same day. And so there was a lot of uh assumptions that it could have been the same threat actor, that this was a coordinated attack. I'm gonna say, based on what I've seen, I don't think they're linked. I'm not saying it's not possible, but essentially what happened is um Polymarket was also compromised last week. It actually came through one of their providers, their tools, debugbear.

Jenn GileDebug Bear had been compromised through its DNS server records. Uh rather, its DNS server records were compromised through its domain registrar, which to all my friends who say it's always DNS, it's always it was DNS.

Paul McCartyUm there's really not always DNS, but it's always DNS.

Jenn GileYeah, exactly. If it's not that, it's secrets. Um yeah, we don't have any other information. This is quite the void of what's happened. I looked around before our um recording today, couldn't find any statements from debug bear other than the initial, hey, we're going through an incident uh thing on X. I don't know, Paul. Have you seen any more than that?

Paul McCartyI mean, I think it's you know, we're we're not gonna go into a lot of detail just because of time, but I I think this debug bear attack is a sign of things to come. Now, this is this has been swept into the dustbin of history because it's only 6.7 million dollars or whatever ridiculous, you know, when you're talking about you know, the the the Binance hack at six billion, this is nothing compared to that. But injection of these malicious JavaScript components from a trusted third-party vendor.

Paul McCartyLet me tell everybody, this attack is gonna be, you know, if I'm gonna call it the the 2027, like this is gonna become much more popular in 2027. Um, what do you think the bad guys are doing right now with all those that hall of credentials they've they've gotten from ghost actions and team PCP and all that? What do you think they're doing with it right now? Right? They're gonna use it to build stuff like this in all kinds of small little companies all over the place. Um, and you know, it's it's gonna be crazy.

Paul McCartyInjected JavaScript from a trusted third party. I'm calling it now.

Jenn GileI have people reach out all the time, and I had an exchange earlier this week about you know the overlap of our area of research and our threat until with ransomware and things like that. And I was telling them, you know, we don't really see a lot of ransomware. And here's more kind of like the reason that I would say that is, as opposed to fact. And Paul, I think you kind of went there with this, is sure, there's a lot of data being collected that you could cash in on and try to get paid by each company that you stole it from.

Jenn GileBut a lot of this stuff can be quickly rotated. You know, if you're stealing credentials, maybe you couldn't get to the data you wanted. And you may get more out of it if you use it as a cascading software supply chain attack, much like what happened with Trivi earlier this year, than if you go and try to extort each victim. And I think that's really you know what we see here.

Paul McCartyYeah, agreed. And I think that you know, we we're gonna say the same thing, which is like AI just makes so much more of this so much more capable. Like something that we're seeing at a very high level is that relatively low maturity threat actors can use AI to build the shape of things. You know, a really good example is ghost action. You know, they're building their payloads and their attacks around the shape of DPRK TTPs. Now, ghost action is not DPRK, but they happen to, for whatever reason, copy the exact payload path name of a fake font file from DPRK to put their payload.

Paul McCartyPayload has nothing to do, it's totally dissimilar from the DPRK one. But my point in bringing all this up is to say low maturity or non-technical threat actors using AI can use the shape of things and then kind of customized their own payloads and their own kind of attack structure around that. And I think we're gonna see that now with these injected, you know, third-party JavaScript. By the way, this attack has been around. I've been dealing with this in WordPress incident response for years, where basically, you know, oh the debug bear thing.

Jenn GileYeah, yeah.

Paul McCartyWell, just yeah, you just like injected injected JavaScript, right? Like you you find you find a WordPress site or a Drupal site that has an upload function and path traversal, and you use those two things together. You pop the the I mean, I literally have been dealing with that for I don't know, 10, 15, I don't know, a long time. And so this idea that now people are gonna take that and they're gonna actually embed that in third party. So they compromise somebody like a post hoc, or they compromise, and I'm not saying postal has been compromised, I'm just saying somebody like that that everybody uses, and then the inject JavaScript into your web app directly, boom.

Paul McCartyYou know, it's like Jeff Kiss.

Jenn GileAll right. Now, uh our last topic today uh would have been the headline for this had it happened at any other time. You and I were working on this, you know, Wednesday, Thursday last week, and then all of a sudden everything went sideways. So uh you were poking around at a compromised package that npm had removed what, ages ago. Uh this predates all blended into the background, who knows?

Jenn GileYeah, it predates um open source malware's existence to tell you how old this is. Not that we're like really old, but you know, uh it gives you an idea. This is not like a current thing. But you were poking around and you said, huh, the shape of this looks like something familiar. So this was an NPM package from a legitimate maintainer where a single version had been compromised with an obfuscated JavaScript payload that looked an awful lot like PollenWriter.

Jenn GileAnd you did some digging. We both kind of played around in packages and GitHub. It was a maintainer who had all of their stuff compromised. And um I'm gonna call this a precursor or a building block to Pollenwriter rather than saying, you know, definitively this was Pollenwriter in May 2025. But it is absolutely the shape of the payload that we see the Pollenwriter campaign force pushing into files, right?

Paul McCartyI'm gonna go further. I'm gonna say this is Pollenwriter.

Jenn GileI'm gonna say this is oh, okay. So what makes it Pollenwriter if it's if it's missing a force push from the dev machine and if it's missing a bunch of other components, what how do we differentiate a piece of malware from a campaign?

Paul McCartyThat's a great question. And I was thinking about this before you said it. All right, so does it have everything that the modern Pollenwriter has? No, it doesn't. And I think that's something else too. Somebody else months ago pointed out um, one of the Isaacs pointed out that hey, we saw Pollenwriter in in September of 2025. And I responded back to that. I was like, no, you didn't. You saw ether hiding, right? And ether hiding is just one component of this whole package and campaign and threat attack structure that has become Pollenwriter, right? So, to your question, it's missing a few things, but I think overall the combination of what is there is absolutely the first version of Pollenwriter.

Jenn GileAnd yeah, that's fair because the way that the malwareware works is very similar. The delivery mechanism is different. This was not a forked repo that you know carried it over.

Paul McCartyBut it wasn't account takeover. I mean, it wasn't a compromise, right? So, and we don't, and because we don't, because it was so long ago, we don't actually have one of the reasons I don't know for sure this is Pollenwriter is because the API I typically call in in GitHub ages out after 90 days, right? So I just can't get this information. So the reality is that there could be force pushes there that we just don't know about because they're gone, they're they're wiped away in the history of the API event log. Um but um the reality is that it was absolutely an account compromise.

Paul McCartySo without the evidence of the force push, we have that component, we just don't know for sure if it was that commit component. So I think taking all those things together, the fact that there's a campaign marker and it's an early campaign marker, right? And just the whole thing kind of blows my mind because, in my worldview, right, like Pollenwriter as a fully fledged thing, you know, and it's evolved a ton since then, really came into existence in in late January or early February 2026.

Paul McCartyThis is a relatively recent thing, as a complete package. Ether hiding existed before, force pushes by DPRK existed before, but putting it all together. Here we go. Here's a package from what May of 2025. And I'm just it's blowing my mind. I'm like, it's it just resets, you know, it's like when we realize that humans have actually been in Australia much longer than we thought they were, right? Because the bridge and all that disappeared and all that like it just it resets the fossil history and the archaeological evidence.

Paul McCartyUm, and so then that means, well, actually, if it existed in May, I can look for some of these things earlier, and we're missing some of these tools that we well.

Jenn GileFor example, one thing you were able to trace back was although the package was from May, some of the infrastructure was even older.

Paul McCartyExactly, exactly. February of 2025.

Jenn GileYeah.

Paul McCartyUm, yeah, so it just it's just, I mean, I'm I'm gonna be honest, I'm a bit verklempt. Like, I it's it's blown my mind a bit. Um, and it's made me question some of those things that you know I I thought were absolute timeline things, pins in a timeline. Uh, clearly they're not.

Jenn GileYeah, well, and this particular developer remained compromised, got re-compromised, not totally clear again because we can't see the total history, but we see in 2026 multiple force pushes, evidence of this developer trying to kick Paul and Ryder out of their repos. So, like the the modern version, we'll call it that. The prehistoric version hit them in 2025. The modern version hit them multiple times this year. So, I my hypothesis, which I think is strongly supported, is this person has been compromised since at least May, probably consistently.

Jenn GileBut that was a like maybe experimental phase. And then the threat actor kind of figured out their playbook and really launched it, you know, in early 2026. And that's the evidence that we see in their repo in 2026.

Paul McCartyYeah, and and you know, as an observation for the for the threat hunters out there and the instant responders, when you find one of these super early events that sits outside of your kind of known timeline is really important, as Jen points out, to really kind of look at that initial uh activity because this is like a lighthouse event. What DPRK was doing with this threat actor, sorry, with this compromised victim, you know, you can you can see how the this is the beginning of something that has now, you know, tens of thousands of of repositories. So being able to tie that back and understanding that kind of evolution, that that's very, very helpful for me, right?

Paul McCartyUm, and and I think it should also be noted for other threat hunters and instant responders. If you get that whole timeline, if you understand how it evolved, and like a real evolutionary chain, there's like little branches, right? I mean, we see Pollen Rider do all kinds of weird stuff. Their obfuscation is constantly kind of evolving and morphing. Um, and right now, for I'm even scared to say this in case they're listening, but right now, the obfuscation mechanism they're using is exposing the initial uh blockchain, the Ethereum blockchain address. So you can now search for the blockchain, and they're using two or three new blockchain Ethereum addresses, and that's not they're not obfuscating that.

Paul McCartySo it's clearly the right there in the pawn rider payload. So um, yeah, it's been interesting.

Jenn GileWell, I'm gonna share one more thing that I haven't even told you about yet. So um one of the GitHub community conversations that we participate in where people are kind of trading, hey, I got compromised, what do I do? Um, I got an alert the other day that someone had shared a whole bunch of um like uh search strings. And I looked at them and I was like, oh, those are really old. Like, don't use those. You're not gonna find the current stuff if you search by those. I kind of gave them some pointers.

Jenn GileUm, but this is like the re-reminder that there's certain things that change frequently and lately have been changing more and more frequently. Those things include the file name, the file ending, and the file location. Don't waste your time setting detections on those unless you're explicitly trying to like see how things are evolving. Um set your detections based on things that are much harder to change.

Jenn GileThe campaign marker, the wallets, etc. You know, it's funny, I was um prepping with our friend Josh from Feebly the other day, because I'm gonna do a webinar with them next month. And he's like, Oh, show me what you're gonna demo. And so I the thing I'm gonna demo is um, you know, you can use our free data to pull up um a sample of a compromised repo, and then use that to go to GitHub, look at the sample, and start, you know, writing your own detections. And so I was showing him how I would do that, and I popped into that repo and I was like, uh, this is not what I documented last week.

Jenn GileUh, since last week, the threat actor had gone in and pushed a new version out with a new file name, file ending of I think it's dot or c I don't even know what that is. It's not a font file. Um, but yeah, it was like a real live example of how had I been searching for the WAF2 example, never would have found it, never would have known that another evolution had happened.

Paul McCartyYeah, if you're looking if to Jen's point, just to go a little bit technically, because she's alluded to it. If you're looking for the WAF2, you're not gonna find the new stuff. And if you're looking just in public slash fonts, you might not find it as well, because now they're burying the fonts directory in other places. Same thing even with the the tasks, uh uh tasks.json now. I've noticed they've started putting that in we're burying it inside of custom areas inside of the repository.

Jenn GileI mean, it makes sense because it takes about two seconds to look at a repo and see, oh, I see a VS Code and a public fonts folder top level of this repo. It's compromised. So they're figuring that out.

Paul McCartyHere's your here's your simplest detection. Does task.json exist? And if so, does it have inside of content run on folder open, right? Yeah, those two. You want to take a look at it.

Jenn GileAnd it might not be malware, but also you probably don't want it. You know, I've got two samples where it's legitimate, but turn that off.

Paul McCartyWell, and also when you're looking at it and you think this is legitimate, like a developer actually did this, and do they really understand what this opens themselves up to? Because they've enabled, they've obviously uh you know, they've re-enabled what Microsoft turned off in February, which is disabling um auto tasks run by default. They're turning it back on and they're you know they're running commands and doing all kinds of stuff. Like you see why tasks, why VS Code tasks is still like DPRK preferred, even like after all this time, it's still delivering for them.

Paul McCartyUm, so you know, DPRK loves Microsoft VS Code.

Jenn GileOkay. Uh, I'm gonna end us on a teaser. I'm very excited. Uh, we don't do guests very often on this podcast, but we uh have a guest lined up for next week. They are the maintainer of two major projects that got compromised this year. And so uh Jared Ray is gonna come on and talk about his experience with uh Kiwi and Cashable getting compromised while his account was compromised, and what he's done as a you know professional maintainer since then.

Jenn GileSo I think it's gonna be super interesting. It's gonna be different than what we normally talk about, but I really believe the community needs to be exposed to what developers go through when uh a project of theirs is compromised.

Paul McCartyTotally agree. I'm looking forward to it. And I'm hoping that it goes a little bit longer than our normal ones, like today. Today's gone long, but yeah, I think it'll go a little long.

Jenn GileAnyway, uh good talk. Have a great weekend, everyone.

Paul McCartyThanks, guys. Appreciate you listening. Cheers.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. React2STEAL hits Node.js/React shops, MSFT official agent-framework compromised

    Nov 18, 202642 minAudio

    This week we talked about: React2STEAL: An automated hacking framework runs entirely on GitHub Actions inside 45 compromised GitHub repositories. It scans for React2Shell and commits stolen credentials back to those...

    Credit the people on this episode

  2. Keyv and Cachable maintainer talks about ChainDrop attack

    Nov 13, 20261 hr 8 minAudio

    This week we had a special guest on the show: Jared Wray, who is the maintainer of several popular open source projects, including Keyv and Cacheable. In August 2026, he was the victim of an account takeover. A threat...

    Credit the people on this episode

  3. Brotli vulnerability, LiteLLM breached via unrevoked key, poisoned prompt downloads malware

    Oct 28, 202631 minAudio

    This week we talked about: Brotli vulnerability: The Brotli decoder bug behind Node2Shell reaches far beyond Node, and it's been overshadowed by the hype around PHP2Shell. Google's Brotli C code is usually vendored into...

    Credit the people on this episode

  4. OVH CLI compromised by Glassworm, compound payloads hide malware, GhostAction copycats PolinRider

    Oct 21, 202642 minAudio

    This week we talked about: OVH CLI compromised by GlassWorm: GlassWorm is back. The Russian threat actor went quiet after its infrastructure was taken down earlier this year, and it resurfaced with malicious versions of...

    Credit the people on this episode

  5. GitHub security improvements, GhostAction phishes OSS maintainers, and DoJ shuts down DPRK laptop farms

    Oct 15, 202627 minAudio

    This week we talked about: GitHub push attribution: GitHub now shows when the account that pushed a commit is different from the one listed as the committer, which makes faked Git history easier to spot. That could help...

    Credit the people on this episode