Blog
/
/
April 17, 2024

Cerber Ransomware: Dissecting the three heads

Cerber ransomware's Linux variant is actively exploiting CVE-2023-22518 in Confluence servers. It uses three UPX-packed C++ payloads: a primary stager, a log checker for environment assessment, and an encryptor that renames files with a .L0CK3D extension.
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
17
Apr 2024

Introduction: Cerber ransomware

Researchers at Cado Security Labs (now part of Darktrace) received reports of the Cerber ransomware being deployed onto servers running the Confluence application via the CVE-2023-22518 exploit. [1] There is a large amount of coverage on the Windows variant, however there is very little about the Linux variant. This blog will discuss an analysis of the Linux variant. 

Cerber emerged and was at the peak of its activity around 2016, and has since only occasional campaigns, most recently targeting the aforementioned Confluence vulnerability. It consists of three highly obfuscated C++ payloads, compiled as a 64-bit Executable and Linkable Format (ELF, the format for executable binary files on Linux) and packed with UPX. UPX is a very common packer used by many threat actors. It allows the actual program code to be stored encoded in the binary, and at runtime extracted into memory and executed (“unpacked”). This is done to prevent software from scanning the payload and detecting the malware.

Pure C++ payloads are becoming less common on Linux, with many threat actors now employing newer programming languages such as Rust or Go. [2] This is likely due to the Cerber payload first being released almost 8 years ago. While it will have certainly received updates, the language and tooling choices are likely to have stuck around for the lifetime of the payload.

Initial access

Cado researchers observed instances of the Cerber ransomware being deployed after a threat actor leveraged CVE-2023-22518 in order to gain access to vulnerable instances of Confluence [3]. It is an improper authorization vulnerability that allows an attacker to reset the Confluence application and create a new administrator account using an unprotected configuration restore endpoint used by the setup wizard.

[19/Mar/2024:15:57:24 +0000] - http-nio-8090-exec-10 13.40.171.234 POST /json/setup-restore.action?synchronous=true HTTP/1.1 302 81796ms - - python-requests/2.31.0 
[19/Mar/2024:15:57:24 +0000] - http-nio-8090-exec-3 13.40.171.234 GET /json/setup-restore-progress.action?taskId= HTTP/1.1 200 108ms 283 - python-requests/2.31.0 

Once an administrator account is created, it can be used to gain code execution by uploading & installing a malicious module via the admin panel. In this case, the Effluence web shell plugin is directly uploaded and installed, which provides a web UI for executing arbitrary commands on the host.

Web Shell recreation
Figure 1: Recreation of installing a web shell on a Confluence instance

The threat actor uses this web shell to download and run the primary Cerber payload. In a default install, the Confluence application is executed as the “confluence” user, a low privilege user. As such, the data the ransomware is able to encrypt is limited to files owned by the confluence user. It will of course succeed in encrypting the datastore for the Confluence application, which can store important information. If it was running as a higher privilege user, it would be able to encrypt more files, as it will attempt to encrypt all files on the system.

Primary payload

Summary of payload:

  • Written in C++, highly obfuscated, and packed with UPX
  • Serves as a stager for further payloads
  • Uses a C2 server at 45[.]145[.]6[.]112 to download and unpack further payloads
  • Deletes itself off disk upon execution

The primary payload is packed with UPX, just like the other payloads. Its main purpose is to set up the environment and grab further payloads in order to run.

Upon execution it unpacks itself and tries to create a file at /var/lock/0init-ld.lo. It is speculated that this was meant to serve as a lock file and prevent duplicate execution of the ransomware, however if the lock file already exists the result is discarded, and execution continues as normal anyway. 

It then connects to the (now defunct) C2 server at 45[.]145[.]6[.]112 and pulls down the secondary payload, a log checker, known internally as agttydck. It does this by doing a simple GET /agttydcki64 request to the server using HTTP and writing the payload body out to /tmp/agttydck.bat. It then executes it with /tmp and ck.log passed as arguments. The execution of the payload is detailed in the next section.

Once the secondary payload has finished executing, the primary payload checks if the log file at /tmp/ck.log it wrote exists. If it does, it then proceeds to delete itself and agttydcki64 from the disk. As it is still running in memory, it then downloads the encryptor payload, known internally as agttydcb, and drops it at /tmp/agttydcb.bat. The packing on this payload is more complex. The file command reports it as a DOS executable and the bat extension would imply this as well. However, it does not have the correct magic bytes, and the high entropy of the file suggests that it is potentially encoded or encrypted. Indeed, the primary payload reads it in and then writes out a decoded ELF file back using the same stream, overwriting the content. It is unclear the exact mechanism used to decode agttydcb. The primary payload then executes the decoded agttydcb, the behavior of which is documented in a later section.

2283  openat(AT_FDCWD, "/tmp/agttydcb.bat", O_RDWR) = 4 
2283  read(4, "\353[\254R\333\372\22,\1\251\f\235 'A>\234\33\25E3g\335\0252\344vBg\177\356\321"..., 450560) = 450560 
2283  lseek(4, 0, SEEK_SET)             = 0 
2283  write(4, "\177ELF\2\1\1\0\0\0\0\0\0\0\0\0\2\0>\0\1\0\0\0X\334F\0\0\0\0\0"..., 450560) = 450560 
2283  close(4)                          = 0 

Truncated strace output for the decoding process

Log check payload - agttydck

Summary of payload:

  • Written in C++, highly obfuscated, and packed with UPX
  • Tries to write the phrase “success” to a given file passed in arguments
  • Likely a check for sandboxing, or to check the permission level of the malware on the system

The log checker payload, agttydck, likely serves as a permission checker. It is a very simple payload and was easy to analyze statically despite the obfuscation. Like the other payloads, it is UPX packed.

When run, it concatenates each argument passed to it and delimits with forward slashes in order to obtain a full path. In this case, it is passed /tmp and ck.log, which becomes /tmp/ck.log. It then tries to open this file in write mode, and if it succeeds writes the word “success” and returns 0. If it does not succeed, it returns 1.

cleaned-up routine
Figure 2: Cleaned-up routine that writes out the success phrase

The purpose of this check isn’t exactly clear. It could be to check if the tmp directory is writable and that it can write, which may be a check for if the system is too locked down for the encryptor to work. Given the check is run in a process separate to the primary payload, it could also be an attempt to detect sandboxes that may not handle files correctly, resulting in the primary payload not being told about the file created by the child.

Encryptor - agttydck

Summary of payload:

  • Written in C++, highly obfuscated, and packed with UPX
  • Writes log file /tmp/log.0 on start and /tmp/log.1 on completion, likely for debugging
  • Walks the root directory looking for directories it can encrypt
  • Writes a ransom note to each directory
  • Overwrites all files in directory with their encrypted content and adds a .L0CK3D extension

The encryptor, agttydcb, achieves the goal of the ransomware, which is to encrypt files on the filesystem. Like the other payloads, it is UPX packed and written with heavily obfuscated C++. Upon launch, it deletes itself off disk so as to not leave any artefacts. It then creates a file at /tmp/log.0, but with no content. As it creates a second file at /tmp/log.1 (also with no content) after encryption finishes, it is possible these were debug markers that the attacker mistakenly left in.

The encryptor then spawns a new thread to do the actual encryption. The payload attempts to write a ransom note at /<directory>/read-me3.txt. If it succeeds, it will walk all files in the directory and attempt to encrypt them. If it fails, it moves on to the next directory. The encryptor chooses to pick which directories to encrypt by walking the root file system. For example, it will try to encrypt /usr, and then /var, etc.

Cerber ransom note
Figure 3: Ransom note left by Cerber

When it has identified a file to encrypt, it opens a read-write file stream to the file and reads in the entire file. It is then encrypted in memory before it seeks to the start of the stream and writes the encrypted data, overwriting the file content, and rendering the file fully encrypted. It then renames the file to have the .L0CK3D extension. Rewriting the same file instead of making a new file and deleting the old one is useful on Linux as directories may be set to append only, preventing the outright deletion of files. Rewriting the file may also rewrite the data on the underlying storage, making recovery with advanced forensics also impossible.

2290  openat(AT_FDCWD, "/home/ubuntu/example", O_RDWR) = 6 
2290  read(6, "file content"..., 3691) = 3691 
2290  write(6, "\241\253\270'\10\365?\2\300\304\275=\30B\34\230\254\357\317\242\337UD\266\362\\\210\215\245!\255f"
..., 3691) = 3691 
2290  close(6)                          = 0 
2290  rename("/home/ubuntu/example", "/home/ubuntu/example.L0CK3D") = 0 

Truncated strace of the encryption process

Once this finishes, it tries to delete itself again (which fails as it already deleted itself) and creates /tmp/log.1. It then gracefully exits. Despite the ransom note claiming the files were exfiltrated, Cado researchers did not observe any behavior that showed this.

Conclusion

Cerber is a relatively sophisticated, albeit aging, ransomware payload. While the use of the Confluence vulnerability allows it to compromise a large amount of likely high value systems, often the data it is able to encrypt will be limited to just the confluence data and in well configured systems this will be backed up. This greatly limits the efficacy of the ransomware in extracting money from victims, as there is much less incentive to pay up.

IoCs

The payloads are packed with UPX so will match against existing UPX Yara rules.

Hashes (sha256)

cerber_primary 4ed46b98d047f5ed26553c6f4fded7209933ca9632b998d265870e3557a5cdfe

agttydcb 1849bc76e4f9f09fc6c88d5de1a7cb304f9bc9d338f5a823b7431694457345bd

agttydck ce51278578b1a24c0fc5f8a739265e88f6f8b32632cf31bf7c142571eb22e243

IPs

C2 (Defunct) 45[.]145[.]6[.]112

References

  1. https://confluence.atlassian.com/security/cve-2023-22518-improper-authorization-vulnerability-in-confluence-data-center-and-server-1311473907.html
  1. https://www.proofpoint.com/uk/threat-reference/cerber-ransomware  
  1. https://nvd.nist.gov/vuln/detail/CVE-2023-22518

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

/

/

August 19, 2026

When AI Becomes the Lure: A Fake Gemini Installer Delivers Vidar

Default blog imageDefault blog image

Key takeaways

  • Darktrace observed a customer download a fake Google Gemini installer hosted on Google Colab, resulting in the execution of the Vidar information stealer.
  • Darktrace identified the compromise through behavioral indicators, including suspicious process activity, anomalous network communications, and indicators of credential theft, before autonomously containing the threat.
  • The incident highlights how threat actors are increasingly exploiting trusted platforms and a growing interest in AI tools to distribute malware through seemingly legitimate software acquisition workflows.

The Growing Abuse of Generative AI

As organizations are increasingly adopting generative AI tools into their daily workflows, attackers are adapting their distribution methods accordingly too. As part of their day-to-day work, users are now searching for AI assistants, programming tools, browser extensions, desktop applications, and productivity integrations.

Recent reports have highlighted campaigns that use fake AI software and AI-related installers to distribute malware and steal credentials [1]. Researchers have documented campaigns that exploit fake AI-themed websites and services to distribute information stealers and backdoors [2]. Security researchers have also observed attackers disguising malware as legitimate installers for AI software to increase the likelihood of victim interaction and execution [3].

In July 2026, Darktrace observed one such case within a customer environment in the Europe, Middle East and Africa (EMEA) region, where attackers used a fake generative AI installer to deliver the prolific information stealer Vidar. This incident highlights how threat actors are exploiting interest in AI services to distribute established malware using increasingly convincing social engineering techniques.

How a Fake Gemini Installer Delivered Vidar

Initial Access: From Search Result to Malware Download

Unlike many malware campaigns that begin with a phishing email, this activity appears to have originated from a user searching for and downloading software.

Darktrace first observed unusual activity on the customer network after a suspicious executable file was launched from a user’s Download folder. Further investigation revealed that the file purported to be a Google Gemini installer and was named “Download_Google_Gemini_For_Windows.exe”.

During the initial analysis, it was noted that the top search result for the suspicious filename associated pointed to a file hosted on Google Colab, a cloud-based Jupyter notebook platform, commonly used by developers, researchers, and data scientists to run code and machine learning workloads through a web browser. By leveraging another trusted Google platform, the attacker increased the likelihood that users would perceive the download as legitimate, making the lure more convincing to those searching for Gemini-related software.

Figure 1: The Google Colab page containing a download prompt for the fake Google Gemini installer.

Further investigation of the Google Colab page revealed that the download prompt redirected users to a secondary site, hxxps://micronsoftwares[.]com, which posed as a "Windows Software Hub" download page and offered the fake Gemini installer for download.

Figure 2: The secondary website posing as a "Windows Software Hub" download page, which likely hosted the fake Gemini installer.

While the investigation did not uncover any HTTP or file-download telemetry data that conclusively identified the download source, SSL communication sessions with Google Colab were detected immediately before the suspicious file was executed. The timing of these connections suggests that the user interacted with the Colab resource before being redirected to the secondary site from which the executable was downloaded.

The user was not simply tricked into opening an email attachment; instead, the attacker embedded malicious content into a process many users would consider entirely legitimate: searching for and downloading software associated with a trusted platform.

Weaponizing Trusted Platforms

At the time of review (July 15, 2026), Darktrace's Threat Research team confirmed that the Google Colab page was still active and prompting users to download a ZIP archive containing the binary file.

The archive also appeared to contain a README file instructing users to run the binary file with administrator privileges and add it to their antivirus software’s exception lists. These instructions suggest that the campaign relied heavily on social engineering, convincing users to take actions that would facilitate malware execution and potentially bypass security checks.

The use of a legitimate platform also complicates the user’s decision-making. Downloads associated with a trusted service are often perceived as less suspicious than those hosted on unfamiliar domains. When combined with the branding of a widely used AI tool, the lure becomes even more convincing.

Malware Analysis

Darktrace’s Threat Research team identified the executable file as the information-stealing malware Vidar. Analysis revealed that the binary file was a newer Go-compiled variant that communicated with Telegram-based infrastructure. Darktrace’s researchers also identified dtm[.]kijangturbo88[.]top as a command-and-control (C2) endpoint associated with the activity. While the malware itself was not novel, the lure and delivery mechanism was.

For a deeper look at the information stealer, see Darktrace’s 2023 analysis of Vidar.

Figure 3: Darktrace’s detection of the unusual outbound connection associated with the fake Gemini installer.

Shortly after execution, the process established communications with the external IP address 91.98.98[.]86 via port 443, directly linking the executable to suspicious network activity observed on the device. Subsequent open-source intelligence (OSINT) analysis of the revealed multiple malicious associations [5].

Additional Darktrace detections included unusual SSL activity from the affected device. Analysis of related SSL telemetry identified 91.98.111[.]49 as additional infrastructure associated  with the activity [6].

Subsequent alerts from the customer's Microsoft Defender for Endpoint integration later confirmed activity consistent with the theft of browser credentials and other sensitive data from the affected endpoint.

Taken together, these detections provided a clear picture of the attack, from the execution of a suspicious file and unusual network connections to indicators of C2 activity and credential theft.

Figure 4: Darktrace’s detection of anomalous activity following the execution of the fake Gemini installer, seen in the Model Alert Event Log.

Darktrace's Autonomous Response

Following the detection, Darktrace’s Autonomous Response took immediate containment action, including blocking communication with suspicious external infrastructure, including 91.98.98[.]86, and quarantining the compromised device.

Despite the apparent legitimacy of the activity, with the installer hosted on a trusted platform and resembling a routine software download, Darktrace was able to detect and contain the attack because the device's behavior deviated from its normal pattern.

Figure 5: Automated containment actions implemented by Darktrace's Autonomous Response following the detection of activity associated with the fake Gemini installer.

Conclusion

This investigation highlights how threat actors continue to adapt established malware delivery techniques to emerging technology trends. While the malware itself was not new, the distribution method was. By disguising Vidar as a Google Gemini installer and hosting the malicious content on a trusted platform, the attack aligned its lure with a growing behavioral trend: users actively searching for AI tools and services as part of their day-to-day work.

Although fake installers are not a new phenomenon, the rapid rise of generative AI has created new opportunities for threat actors. Rather than relying solely on traditional delivery methods, attackers can now target users who are actively searching for AI applications. As AI adoption continues to accelerate across enterprise environments, organizations should remain alert to campaigns that exploit this interest through fake applications, malicious websites, manipulated search results, the misuse of trusted platforms, and AI-themed social engineering.

Credit to Rushanth Ramanathan (Cyber Analyst) Joanna Ng (Detection Engineer)

Edited by Ryan Traill (Content Manager)

Appendices

Darktrace Model Detections

  • Security Integration / C2 Activity and Integration Detection
  • Endpoint / New Suspicious Executable Launched
  • Endpoint / Process Connection / Unusual Connection from New Process
  • Anomalous Connection / Rare External SSL Self-Signed
  • Security Integration / High Severity Integration Detection
  • Antigena / Network / Significant Anomaly /  Antigena Significant Security Integration and Network Activity Block

•Antigena / Network / Significant Anomaly /  Antigena Significant Anomaly from Client Block

List of Indicators of Compromise (IoCs)

IoC Type Description
Download_Google_Gemini_For_Windows.exe File Fake Gemini-themed installer observed during the investigation.
GoogleAppInstaller.exe File Related executable identified through endpoint telemetry.
91.98.98[.]86 IP Address External destination contacted by the malicious executable.
91.98.111[.]49 IP Address Related infrastructure identified through SSL certificate pivoting.
dtm[.]kijangturbo88[.]top Domain Command-and-control endpoint identified during malware analysis.
1e13c2c9eac72daf63fd00a9946878949e159ae6ec51b54ec64f942d79d61913 SHA256 Malware sample associated with the fake Gemini installer.

MITRE ATT@CK Mapping


MITRE ATT&CK Mapping Tactic Technique
Initial Access T1204 User Execution
Execution T1204.002 User Execution: Malicious File
Defence Evasion T1036 Masquerading
Credential Access T1555 Credentials from Password Stores
Credential Access T1555.003 Credentials from Web Browsers
Command and Control T1071 Application Layer Protocol
Exfiltration T1041 Exfiltration Over C2 Channel
Continue reading
About the author
Rajendra Rushanth
Cyber Analyst

Blog

/

Network

/

August 11, 2026

When AI Agents Attack: The Case for Behavioral Anomaly Detection

Default blog imageDefault blog image

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.

Continue reading
About the author
Adam Stevens
Senior Director of Product
Your data. Our AI.
Elevate your network security with Darktrace AI