Handler on Duty: Johannes Ullrich
Threat Level: green
Podcast Detail
SANS Stormcast Friday, July 24th, 2026: OpenAI vs. Huggingface; Zimbra Exploited; Notepad++ Abuse; Browser as C2
If you are not able to play the podcast using the player below: Use this direct link to the audio file: https://traffic.libsyn.com/securitypodcast/10022.mp3
My Next Class
Click HERE to learn more about classes Johannes is teaching for SANS
When the "Autonomous Attacker" Is Your Own AI Model
https://isc.sans.edu/diary/When%20the%20%22Autonomous%20Attacker%22%20Is%20Your%20Own%20AI%20Model/33180
Russian State-Supported Cyber Actors Conduct Phishing Campaign Targeting Users of Zimbra
https://www.cisa.gov/news-events/cybersecurity-advisories/aa26-204a
https://cert.gov.ua/article/6318634
https://cybersecuritynews.com/hackers-abuse-notepad-plugins/
Chaos ransomware's msaRAT: Living off the browser to build a covert C2 channel
https://blog.talosintelligence.com/chaos-msarat-living-off-the-browser-to-build-covert-c2-channel/
My Upcoming Classes
https://www.sans.org/profiles/dr-johannes-ullrich
| Application Security: Securing Web Apps, APIs, and Microservices | Online | British Summer Time | Jul 27th - Aug 1st 2026 |
| Application Security: Securing Web Apps, APIs, and Microservices | Las Vegas | Sep 21st - Sep 25th 2026 |
| Network Monitoring and Threat Detection In-Depth | Amsterdam | Oct 12th - Oct 17th 2026 |
| Application Security: Securing Web Apps, APIs, and Microservices | Washington | Dec 14th - Dec 18th 2026 |
| Application Security: Securing Web Apps, APIs, and Microservices | Online | India Standard Time | Mar 15th - Mar 20th 2027 |
Podcast Transcript
Hello and welcome to the Friday, July 24th, 2026 edition of the SANS Internet Storm Center's Stormcast. My name is Johannes Ullrich, recording today from Jacksonville, Florida. This episode is brought to you by the SANS.edu Master's Degree Program in Information Security Engineering. The OpenAI versus Hugging Face incident does still dominate the news and Renato today wrote up as his diary for the day a summary of it with some lessons learned that I think are quite valuable here. So first of all, that if you have some kind of sandbox or so for your agent, it's typically not providing the security isolation that you expect. It's for many agents really just more of a guideline. Secondly, well, there's this assimility problem around guardrails that Hugging Face ran into where their models actually refused to analyze some of the evidence or respond because of guardrails they ran into. So that's an interesting, I think, lesson to take away from this. And then, of course, well, how did it actually get in? What were the vulnerabilities? And here, non-human identities, API keys and stuff like this still sort of played a major role here. So that's something that, of course, a lot of human -initiated exploits and intrusions are leveraging as well. So definitely something that you probably want to restrict and control. A couple times in the past, I've mentioned cross-site scripting vulnerabilities in Zimbra. So just fitting that, Sisa, in conjunction with other cybersecurity agencies, has published an advisory how a Russian state-sponsored actor is using cross-site scripting vulnerabilities to attack Zimbra users. So Zimbra, like all webmail systems, is struggling with cross-site scripting. You have to essentially display HTML email with cascading style sheets inside the user's browser and, of course, keep it properly separated from the rest of the page. There are some tricks like iframes and such that can help, but ultimately this has been a very difficult problem. And this is one example how these kind vulnerabilities are being attacked. In this case, it was the cascading style sheet that triggered the cross -site scripting vulnerability. And the attacker used it to display a username and password dialogue inside the webmail's user interface. So there was really no good way for the user to distinguish it from the normal username and password dialogue. Once the user entered the username and password, then, of course, the credentials were sent to the attacker. They also took a second factor if that was used. So again, remember, not all second factor authentication is phishing resistant. It's essentially kind of phishing what's happening here, even though it happens inside the page. Something like pass keys or some other 502 authentication would not be vulnerable to this particular attack. I have no idea if Zimbra supports it. Haven't really used Zimbra myself in a while. But anyway, so real good example of how these type of attacks are being exploited, even if you're up to date in your patches. Probably a good read if you're using Zimbra or a similar self-hosted webmail system because, well, like I said, they all tend to have very similar vulnerabilities like this. And we got a blog post by the Ukrainian cert with details regarding, well, other probably Russian attacks, at least target attacks against Ukraine that take advantage of Notepad++. Now, in this case, there's actually not a vulnerability being used, even though that's a little bit disputed here. The problem is that an attacker essentially sending a phishing email, tricking a user into expanding a malicious zip file, and then placing a DLL inside the extension folder for Notepad++. While this DLL essentially replicates a normal extension, it does include additional malicious functionality. There is a dispute whether or not a user should be able to replace a DLL like this. However, this is like how these extensions are installed in Notepad++. So I kind of agree here with Notepad++ that this is not really a vulnerability. It's really just how the product is supposed to work. It's not the user actually willingly installing a malicious extension. Like that's what we usually had in other products like Visual Studio Code and such. Not sure if that's happening with Notepad++ here as well. But in this case, no, the user really just unzips the file and the file opens and looks pretty normal, but has the side mission here of actually replacing and injecting that DLL. Well, earlier this week, I talked about how the calendar tool in Office 365 can be abused as a cover channel. Well, there is no shortage in cover channels. The latest here from Cisco is msaRAT. This particular tool is actually based in Rust, so I guess it's more secure, but it also uses the browser in order to connect to the malicious site. So the command control channel uses Chrome's DevTools protocol. It's an API that developer tools can hook into to basically send requests and receive responses via Google Chrome. That, of course, makes it a little bit more difficult to identify this traffic because when you're looking at the processes on your system, well, the traffic originates from Chrome. There's a lot of weird traffic that goes from and to browsers, so it's easy to kind of discard this as yet another odd kind of browser behavior and not actually follow through with it and look at the source that it's actually initiated by the malware. Well, and this is it for today. So thanks for listening. Thanks for liking, recommending, and for subscribing to this podcast, and talk to you again on Monday. Bye.





