Podcast Detail

SANS Stormcast Tuesday, January 20th, 2026: Scans Against LLMs; NTLM Rainbow Table; OOB MSFT Patch

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/9772.mp3

previous
Podcast Logo
Scans Against LLMs; NTLM Rainbow Table; OOB MSFT Patch
00:00

"How many states are there in the United States?"
Attackers are actively scanning for LLMs, fingerprinting them using the query “How many states are there in the United States?”.
https://isc.sans.edu/diary/%22How%20many%20states%20are%20there%20in%20the%20United%20States%3F%22/32618


Closing the Door on Net-NTLMv1: Releasing Rainbow Tables to Accelerate Protocol Deprecation
Mandiant is publicly releasing a comprehensive dataset of Net-NTLMv1 rainbow tables to underscore the urgency of migrating away from this outdated protocol.
https://cloud.google.com/blog/topics/threat-intelligence/net-ntlmv1-deprecation-rainbow-tables


Out-of-band update to address issues observed with the January 2026 Windows security update
Microsoft has identified issues upon installing the January 2026 Windows security update. To address these issues, an out-of-band (OOB) update was released today, January 17, 2026
https://learn.microsoft.com/en-us/windows/release-health/windows-message-center

Podcast Transcript

 Hello and welcome to the Tuesday January 20, 2026
 edition of the SANS Internet Storm Center's Stormcast. My
 name is Johannes Ullrich and today I'm recording from
 Jacksonville, Florida. And this episode is brought to you
 by the SANS.edu graduate certificate program in
 cybersecurity leadership. And the Didier's Honeypot observed an
 attacker hunting for large language models. More and more
 companies and also individuals are running their internal
 large language models and well some of them are exposing them
 to the internet. And this is what attackers are after. A
 couple reasons for that. First of all of course they could
 just use those models instead of the public and potentially
 more costly ones in order to run their queries. They could
 also possibly then exploit additional weaknesses in it.
 In particular if an individual or a company did add their own
 sort of internal knowledge base to the large language
 model they may be able to enumerate that and then figure
 out exactly you know what kind of secrets or so may be stored
 in that particular model. So a couple of possibilities here.
 We don't exactly know what there are after there are a
 couple other reports as well of people finding these scans
 in their logs. But yeah you probably don't want to expose
 these models to the public without any odd occasion.
 That's really just setting yourself up for at least a
 fairly costly compute bill. And Mandy and the part of
 Google of course now did release a rainbow table of
 possible net NTLM version 1 hashes. Now this is nothing
 really sort of super groundbreaking or such. We all
 know that NTLM version 1 with the single desks and MD4
 hashing is pretty much flawed. But what they're trying to
 push here is that there's still organizations out there
 that are not sort of aggressively removing this
 from their authentication portfolio. So they want to
 give a penetration testers a tool to easily demonstrate
 that these particular authentication mechanisms are
 flawed and that even complex passwords that may be used
 here are not protecting anybody. They're stating that
 it will take you about 600 dollars worth of resources to
 host this rainbow table. I couldn't really find a good
 reference as to how big it is. Now I took a look at the
 Google Cloud storage. We can download them but it even
 didn't really sort of easily state a total size. I'm not
 sure if I just didn't look the right spot here. I'm not that
 familiar with Google Cloud. But so the 600 dollars may be
 just for storage which of course given current hard disk
 prices may be more expensive today than when they released
 this data. Either way it's fairly obtainable this data
 and should take about 12 hours they say to break any NTLM
 hash that you may run into. And they also outline how to
 obtain those hashes, what the different tools are that you
 have available. And then most importantly they're also
 stating and referencing various resources that help
 you move away from NTLM. And Microsoft this weekend took
 the well somewhat unusual step to release an out-of-band
 update on the weekend that did not fix a security flaw.
 Instead it fixed a problem introduced by last Tuesday
 patch Tuesday. And there are two issues that are being
 addressed here. One is where you had problems setting up
 RDP connections and a second one where some systems
 wouldn't shut down or hibernate. So if you ran into
 these problems in particular with a Windows 11 but the RDP
 issue apparently also affects some versions of Windows 10,
 Windows Server 2025, then you may want to make sure that you
 apply this update. And Lihat Eliao with Migo did document
 the interesting vulnerability in Google's calendar and how
 it works with Google's Gemini AI tools. The problem here is
 that NetHacker may send you a calendar invite. That calendar
 invite will include instructions for Gemini to
 summarize not just this meeting but also future
 meetings and exfiltrate the summary by essentially setting
 up a new calendar invite with the summary as part of the
 payload of this calendar invite. A pretty interesting
 exploit the way it sort of skirts some of the
 countermeasures that Gemini has put in place for this type
 of attack. And yet again another example how sort of
 blindly trusting these AI tools can easily get you into
 trouble. This flaw as far as I know and I've read hadn't been
 exploited yet but it's actually sort of in hindsight
 at least a reasonable straightforward flaw that
 probably also exists in numerous similar tools from
 other vendors. Well and this is it for today so thanks for
 listening and thanks for liking thanks for subscribing
 to this podcast and talk to you again tomorrow. Bye!