With the rise of AI assistance features in an increasing number of products, we
have begun to focus some of our research efforts on refining our internal
detection and testing guidelines for LLMs by taking a brief look at the new AI
integrations we discover.
Alongside the rise of applications with LLM integrations, an increasing number
of customers come to ERNW to specifically assess AI applications. Our colleagues
Florian Grunow and
Hannes Mohr analyzed the novel attack
vectors that emerged and presented the results at
TROOPERS24 already.
during a recent Red Teaming engagement Marius Walter from
ERNW found a command injection issue in Progress (Kemp)
LoadMaster. It was registered as
CVE-2024-7591 and scores a
CVSS of 10.0.
The vendor already has patches out, make sure to apply them as this is a high
severe issue. You can find the official announcement and the patch references on
the
official support page.
Marius will follow up with a technical blog post on this issue once we think
everybody had a realistic chance of applying the patches.
This is the first blog post in a series about issues we think are currently
relevant in the field of AI-Security. The intention is not to get full coverage
of the topic, but to point out things that seem practical and relevant. We will
base some of our statements on lab setups and real-life examples. The technology
that we will focus on is chat bots based on generative AI, mainly OpenAI’s
ChatGPT. Right now, this specific application of AI in the wild seems to be the
best way to demonstrate issues and pitfalls when it comes to IT security.
In 2021, ERNW collaborated with
Hochschule Mannheim for their CEP (Cyber Security
Entwicklungsprojekt) to build an auditing framework for testing operating system
configurations against security procedures. This project is part of the
education program of the university to give the students the chance to utilize
the knowledge gained throughout the first semesters in a real world project.
ERNW posed as the fictitious customer, providing a requirements document and
regular meetings with all project groups for feedback. We planned to process and
adapt the results for an open source auditing framework. Unfortunately, we were
not able to finish this project yet, but we think the students should get some
attention for their work independent from our side. So here is a short summary
of what the students created and the corresponding repositories.
I am going to disclose two bug classes I found a while ago in CheckPoint R77.30:
Two buffer overflows in the username (no shit) and HTTP method of a request to
the administrative UI pre-auth and some interesting injections into the TCL web
interface.
Let’s start with the TCL part. The web interface reacted pretty weird when a
payload contained a colon. Diving deeper into this it became clear that a colon
would actually cause an error from the TCL interpreter. By going down this
rabbit hole and learning some TCL (:D) you could see that injecting a colon
breaks some part of the application code, probably because colons are control
characters in TCL e.g. preceding a global variable in TCL (::MyVar) or
separating namespaces.
It is possible to spoof the URLs that Plume will open to arbitrary locations
because of how Plume parses URLs. The preview of an URL in a tweet will show the
complete (at least the host name and the first few chars of the URL) but
shortened URL. However, if the URL contains a semicolon (;) the URL that will be
opened is the part after the semicolon.
An attacker can make use of this behavior by specifying a URL like the following
in a Tweet or direct message:
IoT is everywhere right now and there are a lot of products out there. I have
been looking at an IP Gateway lately and found some serious issues. The
Busch-Welcome IP-Gateway from Busch-Jaeger
is one of the devices that bridges the gap between sensors and actors in your
smart home and the network/Internet. It enables the communication to a door
control system that implements various smart home functions. The device itself
is offering an HTTP service to configure it, which is protected by a username
and password. Some folks even actually expose the device and its login to the
Internet. I tried to configure one of these lately and stumbled upon some
security issues that I would like to discuss in this blog post.
Birk an me basically fully disclosed a 0day in
Squirrelmail yesterday. This is a short Q&A to
answer the most common questions about the issue to calm you all down a little
bit. 😉
What is the punchline, what do I need to know?
An attacker able to exploit this vulnerability can extract files of the server
the application is running on. This may include configuration files, log files
and additionally all files that are readable for all users on the system. This
issue is post-authentication. That means an attacker would need valid
credentials for the application to log in or needs to exploit an additional
vulnerability of which we are not aware of at this point of time.
Some of our Troopers had the chance to visit HITCON conference in Taiwan this
year. There are two main events: HITCON Pacific, which is aimed more at
corporate attendees and HITCON CMT, the community edition, which aims at
students and the general Infosec community. HITCON is the biggest security
conference in Taiwan.
The venue for the event is the Academia Sinica, one of the most important
academic institution in the Republic of China and was founded in 1928 to promote
and undertake scholarly research in sciences and humanities. The conference had
three usual tracks and one special track that was free to use for the public for
demos, presentations and discussions.
Niklaus, Manuel and me had a great time speaking about one of the latest Tablet
PCs from DPRK at
33C3 this year.
Our work on
RedStar OS from last year
revealed a nasty watermarking mechanism that can be used to track the origin and
distribution path of media files in North Korea. We have seen some interesting
dead code in some of RedStar’s binaries that indicated a more sophisticated
mechanism to control the distribution of media files. We got hands on a Tablet
PC called “Ul-lim” that implemented this advanced control mechanism.