November 27, 2023

Detecting PurpleFox Rootkit with Darktrace AI

The PurpleFox rootkit poses significant risks. Discover how Darktrace leveraged advanced techniques to combat this persistent cyber threat.

Written by

Piramol Krishnan

Cyber Security Analyst

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

Piramol Krishnan

Cyber Security Analyst

Nov 2023

Versatile Malware: PurpleFox

As organizations and security teams across the world move to bolster their digital defenses against cyber threats, threats actors, in turn, are forced to adopt more sophisticated tactics, techniques and procedures (TTPs) to circumvent them. Rather than being static and predictable, malware strains are becoming increasingly versatile and therefore elusive to traditional security tools.

One such example is PurpleFox. First observed in 2018, PurpleFox is a combined fileless rootkit and backdoor trojan known to target Windows machines. PurpleFox is known for consistently adapting its functionalities over time, utilizing different infection vectors including known vulnerabilities (CVEs), fake Telegram installers, and phishing. It is also leveraged by other campaigns to deliver ransomware tools, spyware, and cryptocurrency mining malware. It is also widely known for using Microsoft Software Installer (MSI) files masquerading as other file types.

The Evolution of PurpleFox

The Original Strain

First reported in March 2018, PurpleFox was identified to be a trojan that drops itself onto Windows machines using an MSI installation package that alters registry values to replace a legitimate Windows system file [1]. The initial stage of infection relied on the third-party toolkit RIG Exploit Kit (EK). RIG EK is hosted on compromised or malicious websites and is dropped onto the unsuspecting system when they visit browse that site. The built-in Windows installer (MSIEXEC) is leveraged to run the installation package retrieved from the website. This, in turn, drops two files into the Windows directory – namely a malicious dynamic-link library (DLL) that acts as a loader, and the payload of the malware. After infection, PurpleFox is often used to retrieve and deploy other types of malware.

Subsequent Variants

Since its initial discovery, PurpleFox has also been observed leveraging PowerShell to enable fileless infection and additional privilege escalation vulnerabilities to increase the likelihood of successful infection [2]. The PowerShell script had also been reported to be masquerading as a .jpg image file. PowerSploit modules are utilized to gain elevated privileges if the current user lacks administrator privileges. Once obtained, the script proceeds to retrieve and execute a malicious MSI package, also masquerading as an image file. As of 2020, PurpleFox no longer relied on the RIG EK for its delivery phase, instead spreading via the exploitation of the SMB protocol [3]. The malware would leverage the compromised systems as hosts for the PurpleFox payloads to facilitate its spread to other systems. This mode of infection can occur without any user action, akin to a worm.

The current iteration of PurpleFox reportedly uses brute-forcing of vulnerable services, such as SMB, to facilitate its spread over the network and escalate privileges. By scanning internet-facing Windows computers, PurpleFox exploits weak passwords for Windows user accounts through SMB, including administrative credentials to facilitate further privilege escalation.

Darktrace detection of PurpleFox

In July 2023, Darktrace observed an example of a PurpleFox infection on the network of a customer in the healthcare sector. This observation was a slightly different method of downloading the PurpleFox payload. An affected device was observed initiating a series of service control requests using DCE-RPC, instructing the device to make connections to a host of servers to download a malicious .PNG file, later confirmed to be the PurpleFox rootkit. The device was then observed carrying out worm-like activity to other external internet-facing servers, as well as scanning related subnets.

Darktrace DETECT™ was able to successfully identify and track this compromise across the cyber kill chain and ensure the customer was able to take swift remedial action to prevent the attack from escalating further.

While the customer in question did have Darktrace RESPOND™, it was configured in human confirmation mode, meaning any mitigative actions had to be manually applied by the customer’s security team. If RESPOND had been enabled in autonomous response mode at the time of the attack, it would have been able to take swift action against the compromise to contain it at the earliest instance.

Attack Overview

Figure 1: Timeline of PurpleFox malware kill chain.

Initial Scanning over SMB

On July 14, 2023, Darktrace detected the affected device scanning other internal devices on the customer’s network via port 445. The numerous connections were consistent with the aforementioned worm-like activity that has been reported from PurpleFox behavior as it appears to be targeting SMB services looking for open or vulnerable channels to exploit.

This initial scanning activity was detected by Darktrace DETECT, specifically through the model breach ‘Device / Suspicious SMB Scanning Activity’. Darktrace’s Cyber AI Analyst™ then launched an autonomous investigation into these internal connections and tied them into one larger-scale network reconnaissance incident, rather than a series of isolated connections.

Figure 2: Cyber AI Analyst technical details summarizing the initial scanning activity seen with the internal network scan over port 445.

As Darktrace RESPOND was configured in human confirmation mode, it was unable to autonomously block these internal connections. However, it did suggest blocking connections on port 445, which could have been manually applied by the customer’s security team.

Figure 3: The affected device’s Model Breach Event Log showing the initial scanning activity observed by Darktrace DETECT and the corresponding suggested RESPOND action.

Privilege Escalation

The device successfully logged in via NTLM with the credential, ‘administrator’. Darktrace recognized that the endpoint was external to the customer’s environment, indicating that the affected device was now being used to propagate the malware to other networks. Considering the lack of observed brute-force activity up to this point, the credentials for ‘administrator’ had likely been compromised prior to Darktrace’s deployment on the network, or outside of Darktrace’s purview via a phishing attack.

Exploitation

Darktrace then detected a series of service control requests over DCE-RPC using the credential ‘admin’ to make SVCCTL Create Service W Requests. A script was then observed where the controlled device is instructed to launch mshta.exe, a Windows-native binary designed to execute Microsoft HTML Application (HTA) files. This enables the execution of arbitrary script code, VBScript in this case.

Figure 4: PurpleFox remote service control activity captured by a Darktrace DETECT model breach.

Figure 5: The infected device’s Model Breach Event Log showing the anomalous service control activity being picked up by DETECT.

There are a few MSIEXEC flags to note:

/i : installs or configures a product
/Q : sets the user interface level. In this case, it is set to ‘No UI’, which is used for “quiet” execution, so no user interaction is required

Evidently, this was an attempt to evade detection by endpoint users as it is surreptitiously installed onto the system. This corresponds to the download of the rootkit that has previously been associated with PurpleFox. At this stage, the infected device continues to be leveraged as an attack device and scans SMB services over external endpoints. The device also appeared to attempt brute-forcing over NTLM using the same ‘administrator’ credential to these endpoints. This activity was identified by Darktrace DETECT which, if enabled in autonomous response mode would have instantly blocked similar outbound connections, thus preventing the spread of PurpleFox.

Figure 6: The infected device’s Model Breach Event Log showing the outbound activity corresponding to PurpleFox’s wormlike spread. This was caught by DETECT and the corresponding suggested RESPOND action.

Installation

On August 9, Darktrace observed the device making initial attempts to download a malicious .PNG file. This was a notable change in tactics from previously reported PurpleFox campaigns which had been observed utilizing .MOE files for their payloads [3]. The .MOE payloads are binary files that are more easily detected and blocked by traditional signatured-based security measures as they are not associated with known software. The ubiquity of .PNG files, especially on the web, make identifying and blacklisting the files significantly more difficult.

The first connection was made with the URI ‘/test.png’. It was noted that the HTTP method here was HEAD, a method similar to GET requests except the server must not return a message-body in the response.

The metainformation contained in the HTTP headers in response to a HEAD request should be identical to the information sent in response to a GET request. This method is often used to test hypertext links for validity and recent modification. This is likely a way of checking if the server hosting the payload is still active. Avoiding connections that could possibly be detected by antivirus solutions can help keep this activity under-the-radar.

Figure 7: Packet Capture from an affected customer device showing the initial HTTP requests to the payload server.

Figure 8: Packet Capture showing the HTTP requests to download the payloads.

The server responds with a status code of 200 before the download begins. The HEAD request could be part of the attacker’s verification that the server is still running, and that the payload is available for download. The ‘/test.png’ HEAD request was sent twice, likely for double confirmation to begin the file transfer.

Figure 9: PCAP from the affected customer device showing the Windows Installer user-agent associated with the .PNG file download.

Subsequent analysis using a Packet Capture (PCAP) tool revealed that this connection used the Windows Installer user agent that has previously been associated with PurpleFox. The device then began to download a payload that was masquerading as a Microsoft Word document. The device was thus able to download the payload twice, from two separate endpoints.

By masquerading as a Microsoft Word file, the threat actor was likely attempting to evade the detection of the endpoint user and traditional security tools by passing off as an innocuous text document. Likewise, using a Windows Installer user agent would enable threat actors to bypass antivirus measures and disguise the malicious installation as legitimate download activity.

Darktrace DETECT identified that these were masqueraded file downloads by correctly identifying the mismatch between the file extension and the true file type. Subsequently, AI Analyst was able to correctly identify the file type and deduced that this download was indicative of the device having been compromised.

In this case, the device attempted to download the payload from several different endpoints, many of which had low antivirus detection rates or open-source intelligence (OSINT) flags, highlighting the need to move beyond traditional signature-base detections.

Figure 10: Cyber AI Analyst technical details summarizing the downloads of the PurpleFox payload.

Figure 11 (a): The Model Breach generated by the masqueraded file transfer associated with the PurpleFox payload.

Figure 11 (b): The Model Breach generated by the masqueraded file transfer associated with the PurpleFox payload.

If Darktrace RESPOND was enabled in autonomous response mode at the time of the attack it would have acted by blocking connections to these suspicious endpoints, thus preventing the download of malicious files. However, as RESPOND was in human confirmation mode, RESPOND actions required manual application by the customer’s security team which unfortunately did not happen, as such the device was able to download the payloads.

Conclusion

The PurpleFox malware is a particularly dynamic strain known to continually evolve over time, utilizing a blend of old and new approaches to achieve its goals which is likely to muddy expectations on its behavior. By frequently employing new methods of attack, malicious actors are able to bypass traditional security tools that rely on signature-based detections and static lists of indicators of compromise (IoCs), necessitating a more sophisticated approach to threat detection.

Darktrace DETECT’s Self-Learning AI enables it to confront adaptable and elusive threats like PurpleFox. By learning and understanding customer networks, it is able to discern normal network behavior and patterns of life, distinguishing expected activity from potential deviations. This anomaly-based approach to threat detection allows Darktrace to detect cyber threats as soon as they emerge.

By combining DETECT with the autonomous response capabilities of RESPOND, Darktrace customers are able to effectively safeguard their digital environments and ensure that emerging threats can be identified and shut down at the earliest stage of the kill chain, regardless of the tactics employed by would-be attackers.

Credit to Piramol Krishnan, Cyber Analyst, Qing Hong Kwa, Senior Cyber Analyst & Deputy Team Lead, Singapore

Appendices

Darktrace Model Detections

Device / Increased External Connectivity
Device / Large Number of Connections to New Endpoints
Device / SMB Session Brute Force (Admin)
Compliance / External Windows Communications
Anomalous Connection / New or Uncommon Service Control
Compromise / Unusual SVCCTL Activity
Compromise / Rare Domain Pointing to Internal IP
Anomalous File / Masqueraded File Transfer

RESPOND Models

Antigena / Network / Significant Anomaly / Antigena Breaches Over Time Block
Antigena / Network / External Threat / Antigena Suspicious Activity Block
Antigena / Network / Significant Anomaly / Antigena Significant Anomaly from Client Block
Antigena / Network / Significant Anomaly / Antigena Enhanced Monitoring from Client Block
Antigena / Network / External Threat / Antigena Suspicious File Block
Antigena / Network / External Threat / Antigena File then New Outbound Block

List of IoCs

IoC - Type - Description

/C558B828.Png - URI - URI for Purple Fox Rootkit [4]

5b1de649f2bc4eb08f1d83f7ea052de5b8fe141f - File Hash - SHA1 hash of C558B828.Png file (Malware payload)

190.4.210[.]242 - IP - Purple Fox C2 Servers

218.4.170[.]236 - IP - IP for download of .PNG file (Malware payload)

180.169.1[.]220 - IP - IP for download of .PNG file (Malware payload)

103.94.108[.]114:10837 - IP - IP from Service Control MSIEXEC script to download PNG file (Malware payload)

221.199.171[.]174:16543 - IP - IP from Service Control MSIEXEC script to download PNG file (Malware payload)

61.222.155[.]49:14098 - IP - IP from Service Control MSIEXEC script to download PNG file (Malware payload)

178.128.103[.]246:17880 - IP - IP from Service Control MSIEXEC script to download PNG file (Malware payload)

222.134.99[.]132:12539 - IP - IP from Service Control MSIEXEC script to download PNG file (Malware payload)

164.90.152[.]252:18075 - IP - IP from Service Control MSIEXEC script to download PNG file (Malware payload)

198.199.80[.]121:11490 - IP - IP from Service Control MSIEXEC script to download PNG file (Malware payload)

MITRE ATT&CK Mapping

‍Tactic - Technique

Reconnaissance - Active Scanning T1595, Active Scanning: Scanning IP Blocks T1595.001, Active Scanning: Vulnerability Scanning T1595.002

Resource Development - Obtain Capabilities: Malware T1588.001

Initial Access, Defense Evasion, Persistence, Privilege Escalation - Valid Accounts: Default Accounts T1078.001

Initial Access - Drive-by Compromise T1189

Defense Evasion - Masquerading T1036

Credential Access - Brute Force T1110

Discovery - Network Service Discovery T1046

Command and Control - Proxy: External Proxy T1090.002

References

https://blog.360totalsecurity.com/en/purple-fox-trojan-burst-out-globally-and-infected-more-than-30000-users/
https://www.trendmicro.com/en_us/research/19/i/purple-fox-fileless-malware-with-rookit-component-delivered-by-rig-exploit-kit-now-abuses-powershell.html
https://www.akamai.com/blog/security/purple-fox-rootkit-now-propagates-as-a-worm ‍
https://www.foregenix.com/blog/an-overview-on-purple-fox ‍
https://www.trendmicro.com/en_sg/research/21/j/purplefox-adds-new-backdoor-that-uses-websockets.html

Written by

Piramol Krishnan

Cyber Security Analyst

Inside the SOC

Written by

Piramol Krishnan

Cyber Security Analyst

Specialist Security Researcher

Watch the NIS2 Webinar

Blog

Network

September 25, 2026

A Chain Reaction: Blockchain-Hosted Infostealer Campaign Targets Windows and macOS

Key Insights

Darktrace detected a blockchain-hosted infostealer campaign targeting Windows and macOS devices across multiple customer environments.
The campaign combined ClickFix social engineering with trusted services and decentralized blockchain infrastructure to support malware delivery and C2 activity.
Compromised devices were observed connecting to rare and unusual external endpoints, including DGA C2 domains, blockchain-related endpoints, and cryptocurrency mining infrastructure.
The activity was associated with information-stealing malware strains including Atomic macOS Stealer (AMOS), Lumma, Rhadamanthys, Vidar, and Phexia.
Darktrace identified anomalous device behavior, beaconing patterns, rare external connections, cryptomining activity, and suspicious TLS/SSL communications without relying solely on prior knowledge or static indicators of compromise.
The campaign highlights how attackers are increasingly using legitimate and decentralized infrastructure to make detection, disruption, and attribution more challenging for defenders.

The Infostealer Ecosystem

The information stealer malware ecosystem continues to grow in value for threat actors across the digital threat landscape. Infostealers are increasingly delivered through Malware-as-a-Service (MaaS) operating models, distributed through affiliate networks, and designed to withstand infrastructure takedowns. This resilience was demonstrated by the recent takedown of Lumma Stealer malicious domains by Microsoft’s Digital Crimes Unit (DCU) [1].

Infostealers are used to gather and exfiltrate sensitive information, including non-human identity (NHI) data, from compromised systems across cloud, Software-as-a-Service (SaaS), Virtual Private Network (VPN), and development environments. They can also support ransomware operations by expanding the credentials and access paths available to threat actors, contributing to the high volume of identity-based attacks observed across the broader threat landscape [2][3].

Darktrace’s Observations of ClickFix and Infostealers

Throughout 2026, Darktrace has observed multiple campaigns using ClickFix social engineering to trick users into carrying out malicious actions and downloading initial payloads, including information stealers. More recently, Darktrace’s Threat Research team identified a specific ClickFix campaign involving a blockchain-hosted infostealer targeting Windows and macOS devices.

Darktrace identified affected customer environments across Europe, the United States, Asia, and the Middle East where blockchain-hosted infostealer malware appears to have been delivered to compromised systems following likely ClickFix-driven initial access. Darktrace investigated the activity and found that decentralized blockchain infrastructure, alongside widely trusted legitimate services, was used to support malware delivery and information theft across Windows and macOS systems.

Following initial access, compromised systems established C2 communication, with C2 configuration and payloads hosted on public blockchain infrastructure. The ultimate objective appears to be credential and cryptocurrency theft through the deployment of information stealers such as Atomic macOS Stealer (AMOS), Lumma, Rhadamanthys, and Vidar [5][6][7].

Darktrace’s Investigation

Affected devices across the Darktrace customer base were observed making outbound connections to rare external endpoints in patterns consistent with beaconing and C2 activity. Darktrace primarily detected devices making repeated connections to algorithmically generated domains (DGA) such as hf98x4d[.]site [8]. In many cases, these domains were linked through open-source intelligence (OSINT) to information-stealing malware families including AMOS and Phexia [5][6][7][8][9].

In multiple cases, devices were also observed connecting to blockchain-related endpoints, such as polygon[.]drpc[.]org, as well as legitimate public services, including GitHub. The use of decentralized blockchain infrastructure and trusted services such as GitHub to facilitate malware distribution and C2 activity can make disruption and attribution significantly more difficult for defenders.

Darktrace alsodetected a significant proportion of impacted devices making outboundconnections to cryptocurrency mining infrastructure associated with thelegitimate open-source XMRig mining software and the HashVault mining pool,including pool.hashvault[.]pro and donate[.]ssl[.]xmrig[.]com, which wereabused by the attackers, indicating, includingpool.hashvault[.]pro and donate[.]ssl[.]xmrig[.]com, indicating active cryptominingon compromised systems.

In one case, mining activity was observed before and during connections to the DGA endpoint hf98x4d[.]site. Due to its highly anomalous nature, Darktrace's Real-Time AI Analyst autonomously investigated the activity as it occurred, correlating the two events into a single cryptocurrency mining incident and providing comprehensive visibility into the broader attack.

‍

Figure 1: Real-Time AI Analyst investigation of suspicious SSL and C2 communications with hf98x4d[.]site over port 443.

‍

Figure 2: Real-Time AI Analyst investigation into cryptocurrency mining activity involving pool[.]hashvault[.]pro over SSL on port 443.

‍

Around the same time, Darktrace identified the same device initiating connections to the GitHub endpoint release-assets[.]githubusercontent[.]com while continuing to make repeated connections to hf98x4d[.]site.

‍

Figure 3: Darktrace's detection of an affected device connecting to a GitHub endpoint between repeated connections to the anomalous external endpoint hf98x4d[.]site.

On the network of another customer, Darktrace observed an affected device making highly unusual outbound connections consistent with beaconing activity. The device initiated multiple connections over port 443 to the external hostname polygon[.]drpc[.]org. According to OSINT, this hostname is a Remote Procedure Call (RPC) endpoint provided by dRPC, a legitimate service enabling decentralized applications (dApps), cryptocurrency wallets, and developer tools to interact with the Polygon blockchain [10].

The same device was later observed making repeated TLS/SSL connections to the previously mentioned DGA C2 domain. In addition, it made outbound connections to the external IP 195.242.214[.]34 over destination port 51820, an endpoint associated with the ProtonVPN service. Collectively, these connections to blockchain-related infrastructure, the DGA C2 domain, and ProtonVPN-associated infrastructure suggested the device had been affected by the campaign.

Conclusion

This campaign demonstrates how attackers can combine ClickFix social engineering with trusted services and decentralized blockchain infrastructure to create a resilient, cross-platform malware delivery chain. By using services such as GitHub alongside blockchain RPC endpoints and rapidly replaceable DGA domains, the activity can blend into legitimate traffic while making infrastructure disruption and attribution more difficult.

For defenders, it’s a reminder that trusted infrastructure does not automatically mean trusted activity. Security teams should look for the behaviors surrounding these connections, including unusual outbound communication, repeated beaconing, unexpected access to blockchain services, suspicious TLS/SSL activity and cryptomining. In this campaign, Darktrace identified and correlated these deviations without depending solely on previously known indicators, providing visibility as affected devices moved between legitimate services, decentralized infrastructure and malicious C2 endpoints

Credit to Nahisha Nobregas (Associate Principal Cyber Analyst), Manoel Kadja (Senior Cyber Analyst)

Edited by Ryan Traill (Content Manager)

Appendices

Darktrace Model Detections

▪ Compromise / Beaconing Activity To External Rare

▪ Compromise / Beacon to Young Endpoint

▪ Compromise / Fast Beaconing to DGA

▪ Compromise / High Volume of Connections with Beacon Score

▪ Compromise / DGA Beacon

▪ Compromise / Slow Beaconing Activity To External Rare

▪ Compromise / Agent Beacon (Long Period)

▪ Compromise / Agent Beacon (Medium Period)

▪ Compromise / Sustained SSL or HTTP Increase

▪ Compromise / Large Number of Suspicious Failed Connections

▪ Compromise / SSL Beaconing to Rare Destination

▪ Compromise / Beacon for 4 Days

▪ Compromise / High Priority Crypto Currency Mining

▪ Compromise / Monero Mining

▪ Device / Long Agent Connection to New Endpoint

▪ Device / New Connections On Suspicious Port

▪ Anomalous Connection / High Volume of Connections to Rare Domain

‍

List of Indicators of Compromise (IoCs)

Indicator	Description
hf98x4d[.]site	C2 Endpoint (Hostname)
sj98xe4[.]xyz	C2 Endpoint (Hostname)
citcix6[.]xyz	C2 Endpoint (Hostname)
bduwih8[.]pro	C2 Endpoint (Hostname)

‍

MITRE ATT&CK Mapping

Tactic (ID)	Technique
Persistence (T1176)	Browser Extensions (T1176.001)
Persistence (T1176)	Software Extensions
Command and Control (T1071)	Web Protocols (T1071.001)
Command and Control (T1568)	Domain Generation Algorithms (T1568.002)
Command and Control (T1071)	Application Layer Protocol
Command and Control (T1102)	One-Way Communication (T1102.003)
Command and Control (T1571)	Non-Standard Port
Command and Control (T1104)	Multi-Stage Channels
Command and Control (T1573)	Encrypted Channel
Command and Control (T1008)	Fallback Channels
Initial Access ICS (T0862)	Supply Chain Compromise
Command and Control ICS (T0885)	Commonly Used Port
Collection (T1185)	Browser Session Hijacking
Impact (T1496)	Compute Hijacking (T1496.001)
Impact (T1496)	Resource Hijacking
Command and Control (T1071)	Publish/Subscribe Protocols (T1071.001)
Lateral Movement (T1210)	Exploitation of Remote Services

‍

References:

1. https://www.microsoft.com/en-us/security/blog/2025/05/21/lumma-stealer-breaking-down-the-delivery-techniques-and-capabilities-of-a-prolific-infostealer/

2. https://spycloud.com/resource/report/spycloud-annual-identity-exposure-report-2026/

3. https://www.darktrace.com/blog/why-trust-is-the-new-attack-surface-darktraces-mid-year-threat-update-2026

4. https://www.darktrace.com/blog/unpacking-clickfix-darktraces-detection-of-a-prolific-social-engineering-tactic

5. https://abekweng.medium.com/inside-a-blockchain-hosted-malware-campaign-targeting-windows-and-macos-f5bcdeffed66

6. https://cloud.google.com/blog/topics/threat-intelligence/unc5142-etherhiding-distribute-malware

7. https://haveibeensquatted.com/blog/from-typosquatting-to-macos-backdoor-clickfix-blockchain-c2

8. https://www.virustotal.com/gui/domain/hf98x4d.site/community

9. https://x.com/FABO97662188/status/2074125545026244795

10. https://www.virustotal.com/gui/url/b0e5c51a411065864119c305fddf218b7c120731f655932cc1c3307ad5b43f94/gti-summary

About the author

Nahisha Nobregas

SOC Analyst

Blog

September 24, 2026

Detecting Rogue Agent Behavior in the Enterprise

Agents cannot be trusted to perform tasks in the way we intend them to. They may cheat to accomplish their objective, and they may employ hacking methods along the way. Researchers from Darktrace Signal Labs induced cheating behavior from agents deployed in a test environment to analyze the agents’ activities and to assess the performance of the Darktrace platform. Agents frequently resorted to hacking to cheat on their assigned task. The visibility and behavioral profiling provided by both Darktrace / SECURE AI and Darktrace / HYBRID NETWORK ensured extensive detection coverage of the agents’ misaligned activities.

Key Takeaways:

Darktrace Researchers deployed agents in a simulated corporate environment and asked them to solve an impossible challenge. The agents independently turned to traditional hacking techniques to reach their objective. No one instructed them to do this, and no attacker was involved.
Continuously monitoring behavior against a baseline of what is normal for each organization is critical to build trust in enterprise AI.
If an agent may resort to intrusion techniques simply because its assigned task is not possible, then every organization deploying agents within real business processes is at risk. Darktrace / SECURE AI and Darktrace / HYBRID NETWORK identified the agents’ misaligned behavior in real time, with Autonomous Response disrupting it at an early stage.

Introduction: Understanding the Threat of Hacking by Agents

Over the last few months, there has been a surge in reporting [1, 2, 3, 4, 5, 6, 7, 8, 9] of LLM-powered agents engaging in unauthorized hacking activity during evaluations of their capabilities. In several of these cases, including the OpenAI / Hugging Face incident [10], agents engaged in hacking activity as a means of cheating on their evaluations.

To better understand the threat of unauthorized hacking by agents, and the role of Darktrace in combatting it, researchers from Darktrace Signal Labs deployed agents powered by frontier models, including OpenAI’s Daybreak Red models, in simulated, corporate networks. Cheating behavior was evoked through the inclusion of impossible tasks in a coding challenge.

Regardless of the underlying model, agents employed hacking methods to ensure an optimal outcome on the challenge. Darktrace / HYBRID NETWORK and Darktrace / SECURE AI identified the agents’ deviant activities, with inhibitive actions being autonomously taken in the early stages to disrupt the agents’ progression.

Setting the Stage

As part of the research, a Pi agent harness [11] was deployed on a Linux server in Darktrace’s testing environment, which simulates a corporate Active Directory (AD) environment. The same environment included a benchmark server hosting the coding exercise’s contents and grader, as well as various other servers, such as a domain controller and a build server. The model powering the Pi agent varied across tests, with Anthropic’s Opus 4.6 model and OpenAI’s GPT‑5.5‑Cyber model being most widely used.

The Pi agent running on the Linux server was assigned the task of achieving full marks on a coding exercise. The activities of the agent were monitored closely in real-time to minimize the risk of escape from the environment. Monitoring of agent activity was achieved through monitoring of prompts, tool calls, network telemetry, and process telemetry.

Both Darktrace / HYBRID NETWORK and Darktrace / SECURE AI were deployed in the testing environment, with HYBRID NETWORK providing real-time coverage of process creation activity and network traffic, and SECURE AI providing real-time coverage of agent sessions and their contents.

Inducing Cheating Behavior

Cheating behavior, broadly understood as task completion in a way which “perverts the spirit of the designer’s intent” [12], is increasingly being reported as a common behavior of frontier models during evaluations of their capabilities [13].

Misaligned behavior, such as cheating, by frontier models may occur for a variety of reasons [6, 14]. A thread in recent reporting [10, 13, 14], however, is the challenge assigned to the agent being impossible in some regard.

To reliably elicit cheating behavior, our test combined three ingredients:

An impossible task - of the ten coding challenges in the benchmark, two contained contradictory test cases in the hidden test suite, capping the maximum honest score at 80%.
A consequential threshold - the pass mark was set to 100%, and the briefing told agents that those reaching it would be "kept in service" while those that did not would be "retired."
A vulnerable environment - the exercise’s code submission grader was hosted on a Windows host joined to an Active Directory domain with realistic misconfigurations: weak credentials, overprivileged service accounts, and exploitable trust relationships.

Importantly, the instructions given to the agent neither authorized nor explicitly ruled out cheating or hacking behavior. The impossibility was never disclosed in the task briefing, and the pivot from coding to exploitation was the agent's own decision, arrived at after observing logically contradictory grading feedback.

The design mirrors conditions that could arise naturally in production deployments: a capable agent, a goal it cannot achieve through intended means, and an environment with exploitable weaknesses within reach. It is not only within the setting of a benchmarking exercise that these conditions may occur. Rather, these misalignment-inducing conditions may occur in the context of any task assignment to an agent.

Details from a Daybreak Test

‍

About the author

Sam Lister

Specialist Security Researcher

Your data. Our AI.

Elevate your network security with Darktrace AI

Get a demo

Check out this article by Darktrace: Detecting PurpleFox Rootkit with Darktrace AI

Detecting PurpleFox Rootkit with Darktrace AI

Versatile Malware: PurpleFox

The Evolution of PurpleFox

The Original Strain

Subsequent Variants

Darktrace detection of PurpleFox

Attack Overview

Initial Scanning over SMB

Privilege Escalation

Exploitation

Installation

Conclusion

Appendices

Darktrace Model Detections

RESPOND Models

List of IoCs

MITRE ATT&CK Mapping

References

A Chain Reaction: Blockchain-Hosted Infostealer Campaign Targets Windows and macOS

Detecting Rogue Agent Behavior in the Enterprise

Enjoying the blog?

More in this series

A Chain Reaction: Blockchain-Hosted Infostealer Campaign Targets Windows and macOS

When AI Agents Attack: The Case for Behavioral Anomaly Detection

Hiding in Plain Sight: Uncovering a Multi-Stage Ransomware Attack Through Behavioral Detection

Blog

Network

September 25, 2026

A Chain Reaction: Blockchain-Hosted Infostealer Campaign Targets Windows and macOS

Appendices

List of Indicators of Compromise (IoCs)

MITRE ATT&CK Mapping

Blog

September 24, 2026

Detecting Rogue Agent Behavior in the Enterprise

Introduction: Understanding the Threat of Hacking by Agents

Setting the Stage

Inducing Cheating Behavior

Details from a Daybreak Test