Thanks for posting! ~15k lines of code, a lot to poke around in. I was working on a Siemens S7 PLC project with a WINCC HMI for a power plant (the same target of the cyber-weapon) as I listened to the audio book[2] based on this ~12 years ago, entirely changed how I viewed critical industrial infrastructure. One quote from the book that stuck with me was how you can only use a cyber weapon once at full potential, as it’ll either get patched and/or everyone can reverse engineer it to use.
For those not familiar with Stuxnet, it’s a discovered cyber-weapon from 2010 which “reportedly destroyed almost one-fifth of Iran's nuclear centrifuges. ” and “ neither the United States nor Israel has openly admitted responsibility” but likely were the developers [1]
Important to note that the system(s) it infected were non-trivially air-gapped, so it had to do the entirety of it's infectious work without command and control servers or receiving any additional input. It had to be an entirely autonomous process from infection to propagation to execution.
Pretty amazing to have been pulled off seemingly so successfully.
Yes they actually flooded the market with USB drives with Stuxnet. But it was actually too good and started infecting non-target computers. That’s what got a malware researcher to dig in and he saw multiple zero days being used and deduced it was a Mossad and US operation.
Wasn’t it actually not as successful as it could have been? I recall something about it spreading to more machines than it should have, which caused it to be detected earlier than it should have.
Playing with definitions of success in this reply, but, yes, I think the industrial control system detection wasn't quite narrow enough.
Very successful in doing what was intended. Less successful in limiting collateral damage; the collateral damage of which was earlier than intended discovery (or discovery at all).
What lessons have they learned, though, so what's out there now, with 10 years more learning, and now AI assistance?
From what I recall reading, stuxnet itself was found in the wild. Even though it was certainly created for a specific target. That’s how security researchers were able to study it fairly quickly. So the binaries were already out there to pull from other infected systems.
I always enjoyed Kim Zetter's work. Well written and researched, and she doesn't put herself in the story, unlike some other authors. She highlights the practitioners who are actually the ones doing the work.
I always wondered about how feasible the usb drive propagation bit always noted was. Was there ever any evidence that the hardware was already infected at a less scrupulous reseller? I’ve heard of another site in EU that had misbehaving s7. It could have been a reseller that played loose with licensing.
This looks like slop, it's all concatenated into a single file and most probably not based on the actual malware. I'm fairly sure that for example the real one does not include the literal string "Stuxnet" anywhere, like it does here:
The original name given by VirusBlokAda was "Rootkit.Tmphider;"[41]
Symantec, however, called it "W32.Temphid", later changing it to "W32.Stuxnet".[42]
Its current name is derived from a combination of keywords found in the software (".stub" and "mrxnet.sys").[43][44]
And the Wikipedia article explains where the name came from, a combination of ".stub" and "mrxnet.sys".
Not one literal string as it appears several times in this purported "reconstruction". Including as the name for a registry key, in the hex code at the end of an EXE header stub ("REALTEK",0x00,"Stuxnet"), and in a frigging autorun.inf as the program name.
Even if Wikipedia is wrong and that string should appear somewhere in the original binary, whatever LLM they used has really been overdoing it beyond the bounds of realism: "Hey look, it's the REAL STUXNET, you've all read about it, here is the 100% real authentic reverse-engineered source code!"
As for the issues concerning WMI and the Run registry key, I am fully aware of the seriousness of these matters. This is merely a demonstrative demo code, so please just refer to it normally!
AIUI, the successors used a hash of certain system configurations or directory listings to serve as the decryption key for the malicious payload, so that if you found the binary but didn’t have a target system also, it was meaningless.
Directory filtering needs to be fixed, one weird filename or symlink will make it BSOD.
SSDT should probably have a lock. The chance of a race is ~low (higher under heavy sustained workloads) but it's too important to leave to chance.
I'd probably do a rebuild of the directory lists in a separate buffer instead of working in place to avoid alignment fuckups.
Yes I used LLMs, just like I did for all of the other vulns I've found or refined. As you can see from the source, this shit is tedious as hell. Doesn't change the value of knowing what to look/ask for.
Give one of those open models a fresh windows box (not a VM) and tell it to fuck something up, it's fun.
If you're using coding agents for this, it may be worth splitting this up into multiple well arranged modules that tell a coherent story and make it easy to browse, and add explanatory docs based on the various things the LLM has found about each function / type.
Your decompilation threads have all the necessary info in them for this, anyone coming after lacks that foundation and effectively is doing a second inference over the hidden state, assumptions, etc. that your sessions have in them. A simulacrum of a simulacrum in essence is likely to be not particularly good.
Thanks for posting! ~15k lines of code, a lot to poke around in. I was working on a Siemens S7 PLC project with a WINCC HMI for a power plant (the same target of the cyber-weapon) as I listened to the audio book[2] based on this ~12 years ago, entirely changed how I viewed critical industrial infrastructure. One quote from the book that stuck with me was how you can only use a cyber weapon once at full potential, as it’ll either get patched and/or everyone can reverse engineer it to use.
For those not familiar with Stuxnet, it’s a discovered cyber-weapon from 2010 which “reportedly destroyed almost one-fifth of Iran's nuclear centrifuges. ” and “ neither the United States nor Israel has openly admitted responsibility” but likely were the developers [1]
[1-Wikipedia Entry](https://en.wikipedia.org/wiki/Stuxnet)
[2-“Countdown To Zero Day” book if you liked the Wikipedia entry](https://www.audible.com/pd/Countdown-to-Zero-Day-Audiobook/B...)
[3-“Zero Days” movie](https://www.imdb.com/title/tt5446858/)
Important to note that the system(s) it infected were non-trivially air-gapped, so it had to do the entirety of it's infectious work without command and control servers or receiving any additional input. It had to be an entirely autonomous process from infection to propagation to execution.
Pretty amazing to have been pulled off seemingly so successfully.
Yes they actually flooded the market with USB drives with Stuxnet. But it was actually too good and started infecting non-target computers. That’s what got a malware researcher to dig in and he saw multiple zero days being used and deduced it was a Mossad and US operation.
Wasn’t it actually not as successful as it could have been? I recall something about it spreading to more machines than it should have, which caused it to be detected earlier than it should have.
Playing with definitions of success in this reply, but, yes, I think the industrial control system detection wasn't quite narrow enough.
Very successful in doing what was intended. Less successful in limiting collateral damage; the collateral damage of which was earlier than intended discovery (or discovery at all).
What lessons have they learned, though, so what's out there now, with 10 years more learning, and now AI assistance?
Scary thought. We're all pwnt.
Thanks,providing readable code for everyone is inherently a thing of pride.
Where did the binaries come from? Did Iran release them?
From what I recall reading, stuxnet itself was found in the wild. Even though it was certainly created for a specific target. That’s how security researchers were able to study it fairly quickly. So the binaries were already out there to pull from other infected systems.
Are you referring to the government of Iran,
or the (possibly Iranian) sec people they handed the flash drives from the parking lot to?
This is amazingly awesome. Very intriguing and what a great idea in the first place. One of the most important pieces of modern software IMO.
I very much recommend reading "Countdown to Zero Day: Stuxnet and the Launch of the World's First Digital Weapon". Fascinating stuff.
I always enjoyed Kim Zetter's work. Well written and researched, and she doesn't put herself in the story, unlike some other authors. She highlights the practitioners who are actually the ones doing the work.
You might also enjoy "Command and Control" by Eric Schlosser.
I always wondered about how feasible the usb drive propagation bit always noted was. Was there ever any evidence that the hardware was already infected at a less scrupulous reseller? I’ve heard of another site in EU that had misbehaving s7. It could have been a reseller that played loose with licensing.
Given the 'pager attack' supply chain involvement, it can't be ruled out in any other state-sponsored hack / attack.
See also: https://www.schneier.com/blog/archives/2015/03/cisco_shippin...
/seta centrifuge_speed 1000000
g_dwBrainsDestroyedBySlop++;
This looks like slop, it's all concatenated into a single file and most probably not based on the actual malware. I'm fairly sure that for example the real one does not include the literal string "Stuxnet" anywhere, like it does here:
Wikipedia about the origin of the name:Interestinly, it does include the string "myrtus", which was likely it's real internal name.
https://medium.com/@arpit.writes/stuxnet-and-the-birth-of-di...
Yeah... it's reverse engineered. They explained that.
as the other commentator said, it's one thing if it's just variable names (opaque to reverse engineering) but the fact that it is registry keys....
You make a compelling argument.
And the Wikipedia article explains where the name came from, a combination of ".stub" and "mrxnet.sys".
Not one literal string as it appears several times in this purported "reconstruction". Including as the name for a registry key, in the hex code at the end of an EXE header stub ("REALTEK",0x00,"Stuxnet"), and in a frigging autorun.inf as the program name.
Even if Wikipedia is wrong and that string should appear somewhere in the original binary, whatever LLM they used has really been overdoing it beyond the bounds of realism: "Hey look, it's the REAL STUXNET, you've all read about it, here is the 100% real authentic reverse-engineered source code!"
Yeah... you've convinced me. An reverse engineering LLM should/would never invent literal strings in the code.
Astra can reverse engineer binaries now. That's likely where it came from.
But then did it hallucinate that registry key? Or this?
As for the issues concerning WMI and the Run registry key, I am fully aware of the seriousness of these matters. This is merely a demonstrative demo code, so please just refer to it normally!
What?
If you're going to spend tokens on the RE for this, it would've been nice to spend some tokens documenting it or making it easier to navigate
What binary did you start from?
MRxNet and MRxCls.
Somehow, I don't think the original authors of this proprietary code are going to be filing a copyright claim...
It would be really amusing to see, though.
Secrecy aside, there may be some other issues with copywriting a product (allegedly) made by a government.
A NSL seems more likely.
Maybe if this was the real virus, not an AI hallucination.
AIUI, the successors used a hash of certain system configurations or directory listings to serve as the decryption key for the malicious payload, so that if you found the binary but didn’t have a target system also, it was meaningless.
waiting for Pegasus source code
Of course! But it depends. Zero-click is essentially client‑side XSS.
What?
Thanks for posting that
Directory filtering needs to be fixed, one weird filename or symlink will make it BSOD.
SSDT should probably have a lock. The chance of a race is ~low (higher under heavy sustained workloads) but it's too important to leave to chance.
I'd probably do a rebuild of the directory lists in a separate buffer instead of working in place to avoid alignment fuckups.
Yes I used LLMs, just like I did for all of the other vulns I've found or refined. As you can see from the source, this shit is tedious as hell. Doesn't change the value of knowing what to look/ask for.
Give one of those open models a fresh windows box (not a VM) and tell it to fuck something up, it's fun.
If you're using coding agents for this, it may be worth splitting this up into multiple well arranged modules that tell a coherent story and make it easy to browse, and add explanatory docs based on the various things the LLM has found about each function / type.
Waste of tokens IME, better to do it like this and leave that kind of post processing to signature obfuscation.
You don't need exploratory docs, try asking one for complete function signatures of all the Windows APIs used, it's going to spit the docs out.
Your decompilation threads have all the necessary info in them for this, anyone coming after lacks that foundation and effectively is doing a second inference over the hidden state, assumptions, etc. that your sessions have in them. A simulacrum of a simulacrum in essence is likely to be not particularly good.
Thank PR!