Srsly Risky Biz: China's AI-Enabled APT Operations Are Getting Interesting

0:04James Wilson: Hey everyone, I'm James Wilson and welcome to this week's Seriously Risky Biz. This is our cybersecurity policy and intelligence podcast and it is based on the phenomenal newsletter that my colleague Tom Uren writes. You can find Tom's newsletter titled Seriously Risky Business over at our website, risky.biz. You can read the newsletter and also subscribe to it there so it lands in your inbox every week. Tom will be along shortly to talk about the newsletter, but first, big thanks to the sponsor this week, which is Push Security. Push makes an identity security-focused browser extension, which makes a lot of sense when you consider just how much of all of our day-to-day work we now do in the browser.

0:44James Wilson: And earlier this week, I published an interview with VP of Research at Push, Luke Jennings, about how attackers have shifted from authentication phishing into authorization phishing, which is an interesting evolution of their tradecraft. But back to Tom's newsletter. Tom has two topics for this week. The first one is that AI is making Chinese APT operations quite a little bit more interesting at the moment. And in particular, it's sort of moved from let's throw AI at everything and see how it can do the hackathon to actually now doing, I guess you'd say tradecraft the way it should be done but assisted by and force multiplied by AI.

1:21James Wilson: And the second topic is how the US has seemingly decided to throw the kitchen sink at Iranian hackers this week and it's an interesting development with a little bit of a plot twist. But first I'm going to drop you here into the conversation where Tom talks us through you know the number of times that he's written about China and its use of AI in APT operations and espionage in cyber, and how, for the first time, maybe he's actually really quite impressed with the way that they're applying themselves to do, I guess, you say malware creation the way it should be done, but with some interesting novelty added in, thanks to AI.

2:02James Wilson: Enjoy.

2:04Tom Uren: Previous reports I've looked at are kind of people using AI, I would guess in a revolutionary way of let's just throw AI at the problem and let's see what happens. It's, you know, sprinkle magic dust on the problem and we'll get Claude or Gemini or whatever to do hacking, like hands-on keyboard hacking. And that's like newsworthy, it's interesting, but it's not the way serious organizations work. You don't just take, throw away all your software engineering practice and go, let's just replace everything with an AI and get it to do everything. Like that really doesn't make sense, right? And so this report does make sense.

2:48Tom Uren: It talks about a particularly, what looks like a pretty competent Chinese cyber espionage crew who had malware where they were thinking about things like, how do we do things so we don't get caught how do we stay silent so and it talks about how they appear to have incorporated AI into their development practices and it's not to get the AI to be the malware it's not to get the AI to do the hacking it's to get the AI to help them just code faster so instead of having one u-beaut super duper piece of malware and implant they now have seven and they appear to be doing it in a way so that the malware achieves the same things, does the same functions, but just looks very different.

3:40James Wilson: Right. And we'll get to sort of why it's important to have seven of these toolkits up your sleeve in a second. But I just want to drill into sort of the lineage here because you did talk about in the newsletter how you covered this in November. That's back when we were just seeing the, you know, hey, Claude, can you go hack this? There's been stories in the middle though, I'm thinking back to, I think it was even last week when we had the story of what looked to be an autonomous end-to-end attack being launched against Taiwan. Now, in your sense, is this different cyber actors or contractors having different skill sets?

4:13James Wilson: Is this a natural evolution? What does this tell us about the landscape here?

4:18Tom Uren: It feels a bit like let a thousand flowers bloom. Or a thousand monkeys at a typewriter? I think there's room for both approaches, right? And so the previous one I spoke about quite a long time ago, it felt to me like an experiment. We're going to take this, here's one approach, let's try it and see. So in that case, it was the people spent a lot of time building a framework and they let Claude, I believe it was, operate within that framework. And so there was real work there, but it was just open source tools. So it seemed like let's take what we've got, the low equity things that we've got, and see how we can put them together.

4:56Tom Uren: And that's one approach. And I think it makes sense to investigate that approach. In terms of vulnerability discovery, that's turned out to be really worthwhile. It works, right? If you take a harness, have a smart harness that does or gets the models to do the right things in the right order, broken up in the right way that really is very empowering for them it's a massive enabler and i guess the thing about hacking is that you're not operating in isolation there's someone else on the other side of the console or on the other side of the internet that can respond so it's it's a different problem so that you've got different constraints i guess okay so so we're seeing lots

5:42James Wilson: of different experiments coming along it doesn't necessarily indicate you know a shift in approach or a big uplift in maturity. But nevertheless, is that right? Is that the right sort of take?

5:53Tom Uren: Well, I think that there's some actors where the throw the kitchen sink against the wall, you know, just paint by numbers or, you know, very loosely works. It's fine, right? And then there's other actors where you do want to be careful. There's a reason that cyber espionage actors are typically quite cautious and try and act covertly, especially when they're targeting government organizations.

6:20James Wilson: So different contractors or actors trying to achieve different aims is why we see this difference here I think is what I'm taking away from that. Nevertheless now we've got a new player on the scene. This research from Bitdefender calls out Silk Parasite. Before we get to what they've done I'm curious as to how have they attributed so much of this to AI? What did the actual investigation find that made them go oh this is a really interesting use of AI to develop these things?

6:46Tom Uren: Yeah, so there's fingerprints. So the report is somewhat caveated in that it says here these indications, so there was a, you know, leftover pieces of leftover functions, test cases that didn't seem more like AI written than human written. There was examples which are, you know, here's an encryption key and it uses a sort of format, the format that an AI written placeholder would use. The strongest indication, to me at least, was there were two separate pieces of malware that had the exact same architectural choices. And the theory is that they were written with AI assistants from the same high-level specifications document, but the implementation is totally different in the two.

7:42Tom Uren: So one was written in Go, I think. Another was maybe in C. Different programming language. The command and control protocols were totally different. So the implementation was different. It wasn't a port. But the, and so that's not rigorous proof, but it is, this is something that would be very easy to do with AI.

8:02James Wilson: Yeah, it's not the kind of thing that's not, it's not impossible for a human to have done this, but you would beg the question, why would they do it? And of course, there'd be a heck of a lot of effort involved. Okay, so getting to the theme of this use of AI is going to make APT, Chinese APT's actions quite interesting. What is it specifically that makes this interesting? You touched a little bit there on the fact that, okay, they had a spec and they developed it two ways earlier on, you touched on sort of seven different remote access tools being churned out. But why does this make it interesting for us?

8:33Tom Uren: It makes life interesting in the may you live in interesting times kind of curse way, because it makes it harder to tie together activity. So the traditional way for threat intelligence is to tie together activity based on things that you kind of know for certain. They're sometimes called atomic indicators. They're like things that are facts, IP addresses, artifacts in code infrastructure that you used you know mutexes in the software yeah now part of what silk parasite was doing was it was regularly rotating those in the first place but the so that kind of compartments all those indicators but by having different varieties of malware you sort of further split it up and make it harder to associate you know an activity over here in Kazakhstan with an activity in Uzbekistan versus an activity in Tajikistan.

9:33Tom Uren: These malware variants were used, or families I guess, because they weren't really variants, they were just totally different, were used across all those countries. And so it makes it harder for the threat intelligence to then come along and say this is all one actor, here is a, it changes it from a local problem to maybe a global problem.

9:56James Wilson: So interesting insofar as there's some novelty in there, it makes things difficult for traditional attribution, but I still come around to the, aside from making things tricky for attribution and security researchers, what's the big so what of this? What's the macro picture that this starts to create?

10:14Tom Uren: Well, I think that is the big problem. It's hard to respond to something if you don't understand it. and so I don't think it is it makes it impossible I mean this report it's an example of how you can tie together things even though they're taking these mitigation steps but I think for example Salt Typhoon is a global campaign against telecommunications vendors and if that had been cleverly compartmented or broken up into different to make it look like different attackers all of a sudden you've got, oh, well, there's someone attacking the US. That's a problem for the US. There's someone different attacking Australia.

10:57Tom Uren: That's a problem for us. There's someone different in Southeast Asia. And so you don't get a coordinated response.

11:07James Wilson: Do you think that will make them a bit more brazen in their actions or is this just a scale enabler or I guess a longevity play for longer persistence because they're getting less detected? I'm really trying to push for what are the higher ups in the China MSS or other parts of the government that will be overseeing this? What are they saying? Oh, this now allows us to dot, dot, dot. I think that this is an example.

11:34Tom Uren: This may be a bit fatalistic. This is an example of the Red Queen running very, very hard just to stay still. and I think that the entire history of cyber espionage is that it's a continual evolution. The attacker adapts, the defender adjusts. So it's not the end of attribution. I think it means that the techniques that people use have to be more sophisticated. So it's more behavior-based, the timing of operations, exactly when it occurred, all these things become more important. the exact type of malware becomes less important. And I don't think it's, you know, we are now going to hack the planet because we've got seven variants instead of two.

12:19Tom Uren: They were already hacking the planet. But I think it just is another step in the evolution of what goes on. Yeah. But I should say it's an interesting step because this is the first real example I've seen of the really sensible use of malware. So a lot of the examples we've seen previously have been sensational, but I don't want to say nonsensical, but maybe experimental is the right word. Whereas this is, yes, this makes sense. And part of the reason we can see it makes sense is that it's actually a bit harder to detect.

12:54James Wilson: Interesting. Okay. So this to you is evidence that they've dialed in their approach?

12:59Tom Uren: At least some actors. Yeah.

13:00James Wilson: Yeah.

13:01Tom Uren: And like if you were writing malware this is what you do right like it just makes sense and so it's a kind of interesting data point in the i guess learning how to use ai in a way that works right harnessing ai to actually

13:15James Wilson: do something worthwhile okay now speaking of something that may be worthwhile doing maybe not where time will tell the u.s has taken some interesting steps in the last couple of weeks i guess of uh as you say throwing the kitchen sink at iranian hackers now uh what's prompted this and What does the proverbial sink look like in this case? Right.

13:35Tom Uren: So the US has for over a decade now occasionally used a kind of name and shame strategy where they've indicted various state hackers, particularly from China, Russia and Iran in North Korea at times, and just said either in an indictment or in a sanctions announcement, here are all the bad things we allege that you have done and in the last week the u.s launched uh operation economic outcast against iran so this is the next phase in the war on iran that isn't ending as the u.s would like and so in the context of that operation economic outcast it makes sense to like you know throw a few more iranian hackers into the sanctions mix right i think pat made

14:29James Wilson: the point on the show this week that you'd have to imagine someone somewhere and some agents in the US has said, hey, have you got any lead or intention at all of pursuing an Iranian hacker? Let's bump that up to the top of the list and go after this. But there's an interesting twist in this development. This is not just yet another sanction that's only going to matter if they take a trip to somewhere they can be snapped up in. What's the interesting twist that's been buried in all of this.

14:53Tom Uren: Yeah, so like midway through the 5300 word press announcement, they said, and these two hackers have also attacked Iranian interests. And they hacked a particular telecommunications company. So that's interesting, because there's actually a history of intelligence agencies trying to sow dissent and break sow distrust within organisations that they're targeting. So this seems like exactly along those lines. It's interesting because it's a new approach to put it in a press release. And to me, it is like, yeah, that's a really interesting idea. Why would you put it in a Treasury Department press release in the middle, not highlight it?

15:41Tom Uren: Like this is the place that it's most likely to get ignored or discounted by the target audience.

15:48James Wilson: Yeah, it's funny, isn't it? Because it's like on one hand you're making the case that this is designed to be a little surreptitious way to sow some seeds of doubt within the Iranians, but it's causing us to doubt as well. Yeah, yeah. So I think it's a really interesting change in tactic.

16:09Tom Uren: If it is a deliberate change in tactic, it's really interesting. But it just leaves me with so many questions. So I spoke to Hamid Kashfi, who we've had on Between Two Nerds. He is an expert on Iranian cyber espionage. And he kind of agreed that the traditional just say you've sanctions and indictments where you've been hacking a whole lot of US interests, that's not effective. And he said that people actually sometimes get awards for being indicted. You know, job well done. It's proof of you're doing a good job. He said this is interesting because presumably the idea is that the employers, the Ministry of Intelligence and Security or other people in the hacking group don't know that you're doing this.

16:56And so that's, you know, sowing distrust, yeah, that makes sense.

17:01James Wilson: So it used to be a badge of honour. Different ceremony this time? Or do you think this has just missed the mark so much that nothing will come of it?

17:10Tom Uren: It's just a mystery to me. And so it felt like very much right idea. Why would you put it in a press release? It makes no sense.

17:19James Wilson: Where else would it go? I mean, give me a sense.

17:21Tom Uren: Well, the traditional way is that you plant documents in a place or plant evidence in a place the traditional way. In the past, the intelligence agencies have done it is to plant documents. And often those documents have been falsified. So in this case, I think one way to do it would have been to put it in the indictment and say, you know, we've got all this other evidence. Here are the indicators of compromise. And this is why we know as much as they could say. So I think there are alternate venues. You could imagine it could go into a threat intelligence vendors report somehow where it's backed up with actual technical indicators that maybe would make sense.

18:06James Wilson: And Tom, before you go, one thing I wanted to just check with you. What's this I've heard about AI being added to critical infrastructure? And what's going on here and what's your take on this?

18:16Tom Uren: Yeah, so there was a report that came out this week, or maybe last week, that the administration should designate AI as critical infrastructure because it's really important. I think it's right. It is really important. But I also think that mostly you designate things as critical infrastructure when it's important and no one cares about it. Like water is a critical infrastructure sector. and yes, that makes sense because it's underinvested in. Whereas AI is flavour of the month. When the frontier companies complain about a problem like Chinese distillation, the government pays attention and it's front page, is it front page news?

18:58Tom Uren: It's close to front page news. Certainly the government's reaction is front page news. Yes, yes. And so I don't think it's solving a problem that exists. So I think the time to make AI or to designate it as critical infrastructure is when it's boring and no one cares anymore. Gosh, it's hard to imagine what that day will look like. Hopefully it comes.

19:18James Wilson: Yeah, well, the question is, do we get there on a smooth trajectory or is it a whole lot of collateral damage along the way? Let's find out. All right, Tom, great to talk to you. Awesome newsletter again this week. Thanks so much, mate, for dropping by for a chat.

19:31Tom Uren: Thanks, James.

19:46Bye-bye.