ブログ
/
Network
/
August 27, 2024

Decrypting the Matrix: How Darktrace Uncovered a KOK08 Ransomware Attack

In May 2024, a Darktrace customer was affected by KOK08, a ransomware strain commonly used by the Matrix ransomware family. Learn more about the tactics used by this ransomware case, including double extortion, and how Darktrace is able to detect and respond to such threats.
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
Christina Kreza
Cyber Analyst
Default blog image
27
Aug 2024

What is Matrix Ransomware?

Matrix is a ransomware family that first emerged in December 2016, mainly targeting small to medium-sized organizations across the globe in countries including the US, Belgium, Germany, Canada and the UK [1]. Although the reported number of Matrix ransomware attacks has remained relatively low in recent years, it has demonstrated ongoing development and gradual improvements to its tactics, techniques, and procedures (TTPs).

How does Matrix Ransomware work?

In earlier versions, Matrix utilized spam email campaigns, exploited Windows shortcuts, and deployed RIG exploit kits to gain initial access to target networks. However, as the threat landscape changed so did Matrix’s approach. Since 2018, Matrix has primarily shifted to brute-force attacks, targeting weak credentials on Windows machines accessible through firewalls. Attackers often exploit common and default credentials, such as “admin”, “password123”, or other unchanged default settings, particularly on systems with Remote Desktop Protocol (RDP) enabled [2] [3].

Darktrace observation of Matrix Ransomware tactics

In May 2024, Darktrace observed an instance of KOK08 ransomware, a specific strain of the Matrix ransomware family, in which some of these ongoing developments and evolutions were observed. Darktrace detected activity indicative of internal reconnaissance, lateral movement, data encryption and exfiltration, with the affected customer later confirming that credentials used for Virtual Private Network (VPN) access had been compromised and used as the initial attack vector.

Another significant tactic observed by Darktrace in this case was the exfiltration of data following encryption, a hallmark of double extortion. This method is employed by attacks to increase pressure on the targeted organization, demanding ransom not only for the decryption of files but also threatening to release the stolen data if their demands are not met. These stakes are particularly high for public sector entities, like the customer in question, as the exposure of sensitive information could result in severe reputational damage and legal consequences, making the pressure to comply even more intense.

Darktrace’s Coverage of Matrix Ransomware

Internal Reconnaissance and Lateral Movement

On May 23, 2024, Darktrace / NETWORK identified a device on the customer’s network making an unusually large number of internal connections to multiple internal devices. Darktrace recognized that this unusual behavior was indicative of internal scanning activity. The connectivity observed around the time of the incident indicated that the Nmap attack and reconnaissance tool was used, as evidenced by the presence of the URI “/nice ports, /Trinity.txt.bak”.

Although Nmap is a crucial tool for legitimate network administration and troubleshooting, it can also be exploited by malicious actors during the reconnaissance phase of the attack. This is a prime example of a ‘living off the land’ (LOTL) technique, where attackers use legitimate, pre-installed tools to carry out their objectives covertly. Despite this, Darktrace’s Self-Learning AI had been continually monitoring devices across the customers network and was able to identify this activity as a deviation from the device’s typical behavior patterns.

The ‘Device / Attack and Recon Tools’ model alert identifying the active usage of the attack and recon tool, Nmap.
Figure 1: The ‘Device / Attack and Recon Tools’ model alert identifying the active usage of the attack and recon tool, Nmap.
Figure 2: Cyber AI Analyst Investigation into the ‘Scanning of Multiple Devices' incident.

Darktrace subsequently observed a significant number of connection attempts using the RDP protocol on port 3389. As RDP typically requires authentication, multiple connection attempts like this often suggest the use of incorrect username and password combinations.

Given the unusual nature of the observed activity, Darktrace’s Autonomous Response capability would typically have intervened, taking actions such as blocking affected devices from making internal connections on a specific port or restricting connections to a particular device. However, Darktrace was not configured to take autonomous action on the customer’s network, and thus their security team would have had to manually apply any mitigative measures.

Later that day, the same device was observed attempting to connect to another internal location via port 445. This included binding to the server service (srvsvc) endpoint via DCE/RPC with the “NetrShareEnum” operation, which was likely being used to list available SMB shares on a device.

Over the following two days, it became clear that the attackers had compromised additional devices and were actively engaging in lateral movement. Darktrace detected two more devices conducting network scans using Nmap, while other devices were observed making extensive WMI requests to internal systems over DCE/RPC. Darktrace recognized that this activity likely represented a coordinated effort to map the customer’s network and identity further internal devices for exploitation.

Beyond identifying the individual events of the reconnaissance and lateral movement phases of this attack’s kill chain, Darktrace’s Cyber AI Analyst was able to connect and consolidate these activities into one comprehensive incident. This not only provided the customer with an overview of the attack, but also enabled them to track the attack’s progression with clarity.

Furthermore, Cyber AI Analyst added additional incidents and affected devices to the investigation in real-time as the attack unfolded. This dynamic capability ensured that the customer was always informed of the full scope of the attack. The streamlined incident consolidation and real-time updates saved valuable time and resources, enabling quicker, more informed decision-making during a critical response window.

Cyber AI Analyst timeline showing an overview of the scanning related activity, while also connecting the suspicious lateral movement activity.
Figure 3: Cyber AI Analyst timeline showing an overview of the scanning related activity, while also connecting the suspicious lateral movement activity.

File Encryption

On May 28, 2024, another device was observed connecting to another internal location over the SMB filesharing protocol and accessing multiple files with a suspicious extension that had never previously been observed on the network. This activity was a clear sign of ransomware infection, with the ransomware altering the files by adding the “KOK08@QQ[.]COM” email address at the beginning of the filename, followed by a specific pattern of characters. The string consistently followed a pattern of 8 characters (a mix of uppercase and lowercase letters and numbers), followed by a dash, and then another 8 characters. After this, the “.KOK08” extension was appended to each file [1][4].

Cyber AI Analyst Investigation Process for the 'Possible Encryption of Files over SMB' incident.
Figure 4: Cyber AI Analyst Investigation Process for the 'Possible Encryption of Files over SMB' incident.
Cyber AI Analyst Encryption Information identifying the ransomware encryption activity,
Figure 5: Cyber AI Analyst Encryption Information identifying the ransomware encryption activity.

Data Exfiltration

Shortly after the encryption event, another internal device on the network was observed uploading an unusually large amount of data to the rare external endpoint 38.91.107[.]81 via SSH. The timing of this activity strongly suggests that this exfiltration was part of a double extortion strategy. In this scenario, the attacker not only encrypts the target’s files but also threatens to leak the stolen data unless a ransom is paid, leveraging both the need for decryption and the fear of data exposure to maximize pressure on the victim.

The full impact of this double extortion tactic became evident around two months later when a ransomware group claimed possession of the stolen data and threatened to release it publicly. This development suggested that the initial Matrix ransomware attackers may have sold the exfiltrated data to a different group, which was now attempting to monetize it further, highlighting the ongoing risk and potential for exploitation long after the initial attack.

External data being transferred from one of the involved internal devices during and after the encryption took place.
Figure 6: External data being transferred from one of the involved internal devices during and after the encryption took place.

Unfortunately, because Darktrace’s Autonomous Response capability was not enabled at the time, the ransomware attack was able to escalate to the point of data encryption and exfiltration. However, Darktrace’s Security Operations Center (SOC) was still able to support the customer through the Security Operations Support service. This allowed the customer to engage directly with Darktrace’s expert analysts, who provided essential guidance for triaging and investigating the incident. The support from Darktrace’s SOC team not only ensured the customer had the necessary information to remediate the attack but also expedited the entire process, allowing their security team to quickly address the issue without diverting significant resources to the investigation.

Conclusion

In this Matrix ransomware attack on a Darktrace customer in the public sector, malicious actors demonstrated an elevated level of sophistication by leveraging compromised VPN credentials to gain initial access to the target network. Once inside, they exploited trusted tools like Nmap for network scanning and lateral movement to infiltrate deeper into the customer’s environment. The culmination of their efforts was the encryption of files, followed by data exfiltration via SSH, suggesting that Matrix actors were employing double extortion tactics where the attackers not only demanded a ransom for decryption but also threatened to leak sensitive information.

Despite the absence of Darktrace’s Autonomous Response at the time, its anomaly-based approach played a crucial role in detecting the subtle anomalies in device behavior across the network that signalled the compromise, even when malicious activity was disguised as legitimate.  By analyzing these deviations, Darktrace’s Cyber AI Analyst was able to identify and correlate the various stages of the Matrix ransomware attack, constructing a detailed timeline. This enabled the customer to fully understand the extent of the compromise and equipped them with the insights needed to effectively remediate the attack.

Credit to Christina Kreza (Cyber Analyst) and Ryan Traill (Threat Content Lead)

Appendices

Darktrace Model Detections

·       Device / Network Scan

·       Device / Attack and Recon Tools

·       Device / Possible SMB/NTLM Brute Force

·       Device / Suspicious SMB Scanning Activity

·       Device / New or Uncommon SMB Named Pipe

·       Device / Initial Breach Chain Compromise

·       Device / Multiple Lateral Movement Model Breaches

·       Device / Large Number of Model Breaches from Critical Network Device

·       Device / Multiple C2 Model Breaches

·       Device / Lateral Movement and C2 Activity

·       Anomalous Connection / SMB Enumeration

·       Anomalous Connection / New or Uncommon Service Control

·       Anomalous Connection / Multiple Connections to New External TCP Port

·       Anomalous Connection / Data Sent to Rare Domain

·       Anomalous Connection / Uncommon 1 GiB Outbound

·       Unusual Activity / Enhanced Unusual External Data Transfer

·       Unusual Activity / SMB Access Failures

·       Compromise / Ransomware / Suspicious SMB Activity

·       Compromise / Suspicious SSL Activity

List of Indicators of Compromise (IoCs)

·       .KOK08 -  File extension - Extension to encrypted files

·       [KOK08@QQ[.]COM] – Filename pattern – Prefix of the encrypted files

·       38.91.107[.]81 – IP address – Possible exfiltration endpoint

MITRE ATT&CK Mapping

·       Command and control – Application Layer Protocol – T1071

·       Command and control – Web Protocols – T1071.001

·       Credential Access – Password Guessing – T1110.001

·       Discovery – Network Service Scanning – T1046

·       Discovery – File and Directory Discovery – T1083

·       Discovery – Network Share Discovery – T1135

·       Discovery – Remote System Discovery – T1018

·       Exfiltration – Exfiltration Over C2 Channer – T1041

·       Initial Access – Drive-by Compromise – T1189

·       Initial Access – Hardware Additions – T1200

·       Lateral Movement – SMB/Windows Admin Shares – T1021.002

·       Reconnaissance – Scanning IP Blocks – T1595.001

References

[1] https://unit42.paloaltonetworks.com/matrix-ransomware/

[2] https://www.sophos.com/en-us/medialibrary/PDFs/technical-papers/sophoslabs-matrix-report.pdf

[3] https://cyberenso.jp/en/types-of-ransomware/matrix-ransomware/

[4] https://www.pcrisk.com/removal-guides/10728-matrix-ransomware

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
Christina Kreza
Cyber Analyst

More in this series

No items found.

Blog

/

Network

/

September 25, 2026

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

Default blog imageDefault blog image

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

Continue reading
About the author
Nahisha Nobregas
SOC Analyst

Blog

/

AI

/

September 24, 2026

Detecting Rogue Agent Behavior in the Enterprise

Default blog imageDefault blog image

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:

  1. 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%.
  2. 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."
  3. 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

‍

‍

Continue reading
About the author
Sam Lister
Specialist Security Researcher
あなたのデータ × DarktraceのAI
唯一無二のDarktrace AIで、ネットワークセキュリティを次の次元へ