Blog
/
Network
/
May 23, 2023

Darktrace’s Detection of a Hive Ransomware-as-Service

This blog investigates a new strain of ransomware, Hive, a ransomware-as-a-service. Darktrace was able to provide full visibility over the attacks.
Written by
Emily Megan Lim
Cyber 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
Emily Megan Lim
Cyber Analyst
Default blog image
23
May 2023

Update: On January 26, 2023, the Hive ransomware group was dismantled and servers associated with the sale of the ransomware were taken offline following an investigation by the FBI, German law enforcement and the National Crime Agency (NCA). The activity detailed in this blog took place in 2022, whilst the group was still active.‍

RaaS in Cyber Security

The threat of ransomware continues to be a constant concern for security teams across the cyber threat landscape. With the growing popularity of Ransomware-as-a-Service (RaaS), it is becoming more and more accessible for even inexperienced would-be attackers. As a result of this low barrier to entry, the volume of ransomware attacks is expected to increase significantly.

What’s more, RaaS is a highly tailorable market in which buyers can choose from varied kits and features to use in their ransomware deployments meaning attacks will rarely behave the same. To effectively detect and safeguard against these differentiations, it is crucial to implement security measures that put the emphasis on detecting anomalies and focusing on deviations in expected behavior, rather than relying on depreciated indicators of compromise (IoC) lists or playbooks that focus on attack chains unable to keep pace with the increasing speed of ransomware evolution.

In early 2022, Darktrace DETECT/Network™ identified several instances of Hive ransomware on the networks of multiple customers. Using its anomaly-based detection, Darktrace was able to successfully detect the attacks and multiple stages of the kill chain, including command and control (C2) activity, lateral movement, data exfiltration, and ultimately data encryption and the writing of ransom notes.

Hive Ransomware 

Hive ransomware is a relatively new strain that was first observed in the wild in June 2021. It is known to target a variety of industries including healthcare, energy providers, and retailers, and has reportedly attacked over 1,500 organizations, collecting more than USD 100m in ransom payments [1].

Hive is distributed via a RaaS model where its developers update and maintain the code, in return for a percentage of the eventual ransom payment, while users (or affiliates) are given the tools to carry out attacks using a highly sophisticated and complex malware they would otherwise be unable to use. Hive uses typical tactics, techniques and procedures (TTPs) associated with ransomware, though they do vary depending on the Hive affiliate carrying out the attack.

In most cases a double extortion attack is carried out, whereby data is first exfiltrated and then encrypted before a ransom demand is made. This gives attackers extra leverage as victims are at risk of having their sensitive data leaked to the public on websites such as the ‘HiveLeaks’ TOR website.

Attack Timeline

Owing to the highly customizable nature of RaaS, the tactics and methods employed by Hive actors are expected to differ on a case-by-case basis. Nonetheless in the majority of Hive ransomware incidents identified on Darktrace customer environments, Darktrace DETECT observed the following general attack stages and features. This is possibly indicative of the attacks originating from the same threat actor(s) or from a widely sold batch with a particular configuration to a variety of actors.

Figure 1: A typical timeline of a Hive attack observed by Darktrace.

Initial Access 

Although Hive actors are known to gain initial access to networks through multiple different vectors, the two primary methods reported by security researchers are the exploitation of Microsoft Exchange vulnerabilities, or the distribution of phishing emails with malicious attachments [2][3].

In the early stages of one Hive ransomware attack observed on the network of a Darktrace customer, for example, Darktrace detected a device connecting to the rare external location 23.81.246[.]84, with a PowerShell user agent via HTTP. During this connection, the device attempted to download an executable file named “file.exe”. It is possible that the file was initially accessed and delivered via a phishing email; however, as Darktrace/Email was not enabled at the time of the attack, this was outside of Darktrace’s purview. Fortunately, the connection failed the proxy authentication was thus blocked as seen in the packet capture (PCAP) in Figure 2. 

Shortly after this attempted download, the same device started to receive a high volume of incoming SSL connections from a rare external endpoint, namely 146.70.87[.]132. Darktrace logged that this endpoint was using an SSL certificate signed by Go Daddy CA, an easily obtainable and accessible SSL certificate, and that the increase in incoming SSL connections from this endpoint was unusual behavior for this device. 

It is likely that this highly anomalous activity detected by Darktrace indicates when the ransomware attack began, likely initial payload download.  

Darktrace DETECT models:

  • Anomalous Connection / Powershell to Rare External
  • Anomalous Server Activity / New Internet Facing System
Figure 2: PCAP of the HTTP connection to the rare endpoint 23.81.246[.]84 showing the failed proxy authentication.

C2 Beaconing 

Following the successful initial access, Hive actors begin to establish their C2 infrastructure on infected networks through numerous connections to C2 servers, and the download of additional stagers. 

On customer networks infected by Hive ransomware, Darktrace identified devices initiating a high volume of connections to multiple rare endpoints. This very likely represented C2 beaconing to the attacker’s infrastructure. In one particular example, further open-source intelligence (OSINT) investigation revealed that these endpoints were associated with Cobalt Strike.

Darktrace DETECT models:

  • Anomalous Connection / Multiple Connections to New External TCP
  • Anomalous Server Activity / Anomalous External Activity from Critical Network Device
  • Compromise / High Volume of Connections with Beacon Score
  • Compromise / Sustained SSL or HTTP Increase
  • Compromise / Suspicious HTTP Beacons to Dotted Quad 
  • Compromise / SSL or HTTP Beacon
  • Device / Lateral Movement and C2 Activity

Internal Reconnaissance, Lateral Movement and Privilege Escalation

After C2 infrastructure has been established, Hive actors typically begin to uninstall antivirus products in an attempt to remain undetected on the network [3]. They also perform internal reconnaissance to look for vulnerabilities and open channels and attempt to move laterally throughout the network.

Amid the C2 connections, Darktrace was able to detect network scanning activity associated with the attack when a device on one customer network was observed initiating an unusually high volume of connections to other internal devices. A critical network device was also seen writing an executable file “mimikatz.exe” via SMB which appears to be the Mimikatz attack tool commonly used for credential harvesting. 

There were also several detections of lateral movement attempts via RDP and DCE-RPC where the attackers successfully authenticated using an “Administrator” credential. In one instance, a device was also observed performing ITaskScheduler activity. This service is used to remotely control tasks running on machines and is commonly observed as part of malicious lateral movement activity. Darktrace DETECT understood that the above activity represented a deviation from the devices’ normal pattern of behavior and the following models were breached:

Darktrace DETECT models:

  • Anomalous Connection / Anomalous DRSGetNCChanges Operation
  • Anomalous Connection / New or Uncommon Service Control
  • Anomalous Connection / Unusual Admin RDP Session
  • Anomalous Connection / Unusual SMB Version 1 Connectivity
  • Compliance / SMB Drive Write
  • Device / Anomalous ITaskScheduler Activity
  • Device / Attack and Recon Tools
  • Device / Attack and Recon Tools In SMB
  • Device / EXE Files Distributed to Multiple Devices
  • Device / Suspicious Network Scan Activity
  • Device / Increase in New RPC Services
  • User / New Admin Credentials on Server

Data Exfiltration

At this stage of the attack, Hive actors have been known to carry out data exfiltration activity on infected networks using a variety of different methods. The Cybersecurity & Infrastructure Security Agency (CISA) reported that “Hive actors exfiltrate data likely using a combination of Rclone and the cloud storage service Mega[.]nz” [4]. Darktrace DETECT identified an example of this when a device on one customer network was observed making HTTP connections to endpoints related to Mega, including “w.apa.mega.co[.]nz”, with the user agent “rclone/v1.57.0” with at least 3 GiB of data being transferred externally (Figure 3). The same device was also observed transferring at least 3.6 GiB of data via SSL to the rare external IP, 158.51.85[.]157.

Figure 3: A summary of a device’s external connections to multiple endpoints and the respective amounts of data exfiltrated to Mega storage endpoints.

In another case, a device was observed uploading over 16 GiB of data to a rare external endpoint 93.115.27[.]71 over SSH. The endpoint in question was seen in earlier beaconing activity suggesting that this was likely an exfiltration event. 

However, Hive ransomware, like any other RaaS kit, can differ greatly in its techniques and features, and it is important to note that data exfiltration may not always be present in a Hive ransomware attack. In one incident detected by Darktrace, there were no signs of any data leaving the customer environment, indicating data exfiltration was not part of the Hive actor’s objectives.

Darktrace DETECT models:

  • Anomalous Connection / Data Sent to Rare Domain
  • Anomalous Connection / Lots of New Connections
  • Anomalous Connection / Multiple HTTP POSTs to Rare Hostname
  • Anomalous Connection / Suspicious Self-Signed SSL
  • Anomalous Connection / Uncommon 1 GiB Outbound
  • Device / New User Agent and New IP
  • Unusual Activity / Unusual External Data to New Endpoints
  • Unusual Activity / Unusual External Data Transfer
  • Unusual Activity / Enhanced Unusual External Data Transfer

Ransomware Deployment

In the final stage of a typical Hive ransomware attack, the ransomware payload is deployed and begins to encrypt files on infected devices. On one customer network, Darktrace detected several devices connecting to domain controllers (DC) to read a file named “xxx.exe”. Several sources have linked this file name with the Hive ransomware payload [5].

In another example, Darktrace DETECT observed multiple devices downloading the executable files “nua64.exe” and “nua64.dll” from a rare external location, 194.156.90[.]25. OSINT investigation revealed that the files are associated with Hive ransomware.

Figure 4: Security vendor analysis of the malicious file hash [6] associated with Hive ransomware. 

Shortly after the download of this executable, multiple devices were observed performing an unusual amount of file encryption, appending randomly generated strings of characters to file extensions. 

Although it has been reported that earlier versions of Hive ransomware encrypted files with a “.hive” extension [7], Darktrace observed across multiple customers that encrypted files had extensions that were partially-randomized, but consistently 20 characters long, matching the regular expression “[a-zA-Z0-9\-\_]{8}[\-\_]{1}[A-Za-z0-9\-\_]{11}”.

Figure 5: Device Event Log showing SMB reads and writes of encrypted files with a randomly generated extension of 20 characters. 

Following the successful encryption of files, Hive proceeds to drop a ransom note, named “HOW_TO_DECRYPT.txt”, into each affected directory. Typically, the ransom note will contain a link to Hive’s “sales department” and, in the event that exfiltration took place, a link to the “HiveLeaks” site, where attackers threaten to publish exfiltrated data if their demands are not met (Figure 6).  In cases of Hive ransomware detected by Darktrace, multiple devices were observed attempting to contact “HiveLeaks” TOR domains, suggesting that endpoint users had followed links provided to them in ransom notes.

Figure 6: Sample of a Hive ransom note [4].

Examples of file extensions:

  • 36C-AT9-_wm82GvBoCPC
  • 36C-AT9--y6Z1G-RFHDT
  • 36C-AT9-_x2x7FctFJ_q
  • 36C-AT9-_zK16HRC3QiL
  • 8KAIgoDP-wkQ5gnYGhrd
  • kPemi_iF_11GRoa9vb29
  • kPemi_iF_0RERIS1m7x8
  • kPemi_iF_7u7e5zp6enp
  • kPemi_iF_y4u7pB3d3f3
  • U-9Xb0-k__T0U9NJPz-_
  • U-9Xb0-k_6SkA8Njo5pa
  • zm4RoSR1_5HMd_r4a5a9 

Darktrace DETECT models:

  • Anomalous Connection / SMB Enumeration
  • Anomalous Connection / Sustained MIME Type Conversion
  • Anomalous Connection / Unusual Admin SMB Session
  • Anomalous File / Internal / Additional Extension Appended to SMB File
  • Compliance / SMB Drive Write
  • Compromise / Ransomware / Suspicious SMB Activity
  • Compromise / Ransomware / Ransom or Offensive Words Written to SMB
  • Compromise / Ransomware / Possible Ransom Note Write
  • Compromise / High Priority Tor2Web
  • Compromise / Tor2Web
  • Device / EXE Files Distributed to Multiple Devices

Conclusion

As Hive ransomware attacks are carried out by different affiliates using varying deployment kits, the tactics employed tend to vary and new IoCs are regularly identified. Furthermore, in 2022 a new variant of Hive was written using the Rust programming language. This represented a major upgrade to Hive, improving its defense evasion techniques and making it even harder to detect [8]. 

Hive is just one of many RaaS offerings currently on the market, and this market is only expected to grow in usage and diversity of presentations.  As ransomware becomes more accessible and easier to deploy it is essential for organizations to adopt efficient security measures to identify ransomware at the earliest possible stage. 

Darktrace DETECT’s Self-Learning AI understands customer networks and learns the expected patterns of behavior across an organization’s digital estate. Using its anomaly-based detection Darktrace is able to identify emerging threats through the detection of unusual or unexpected behavior, without relying on rules and signatures, or known IoCs. 

Credit to: Emily Megan Lim, Cyber Analyst, Hyeongyung Yeom, Senior Cyber Analyst & Analyst Team Lead.

Appendices

MITRE AT&CK Mapping

Reconnaissance

T1595.001 – Scanning IP Blocks

T1595.002 – Vulnerability Scanning

Resource Development

T1583.006 – Web Services

Initial Access

T1078 – Valid Accounts

T1190 – Exploit Public-Facing Application

T1200 – Hardware Additions

Execution

T1053.005 – Scheduled Task

T1059.001 – PowerShell

Persistence/Privilege Escalation

T1053.005 – Scheduled Task

T1078 – Valid Accounts

Defense Evasion

T1078 – Valid Accounts

T1207 – Rogue Domain Controller

T1550.002 – Pass the Hash

Discovery

T1018 – Remote System Discovery

T1046 – Network Service Discovery

T1083 – File and Directory Discovery

T1135 – Network Share Discovery

Lateral Movement

T1021.001 – Remote Desktop Protocol

T1021.002 – SMB/Windows Admin Shares

T1021.003 – Distributed Component Object Model

T1080 – Taint Shared Content

T1210 – Exploitation of Remote Services

T1550.002 – Pass the Hash

T1570 – Lateral Tool Transfer

Collection

T1185 – Man in the Browser

Command and Control

T1001 – Data Obfuscation

T1071 – Application Layer Protocol

T1071.001 – Web Protocols

T1090.003 – Multi-hop proxy

T1095 – Non-Application Layer Protocol

T1102.003 – One-Way Communication

T1571 – Non-Standard Port

Exfiltration

T1041 – Exfiltration Over C2 Channel

T1567.002 – Exfiltration to Cloud Storage

Impact

T1486 – Data Encrypted for Impact

T1489 – Service Stop

List of IoCs 

23.81.246[.]84 - IP Address - Likely Malicious File Download Endpoint

146.70.87[.]132 - IP Address - Possible Ransomware Endpoint

5.199.162[.]220 - IP Address - C2 Endpoint

23.227.178[.]65 - IP Address - C2 Endpoint

46.166.161[.]68 - IP Address - C2 Endpoint

46.166.161[.]93 - IP Address - C2 Endpoint

93.115.25[.]139 - IP Address - C2 Endpoint

185.150.1117[.]189 - IP Address - C2 Endpoint

192.53.123[.]202 - IP Address - C2 Endpoint

209.133.223[.]164 - IP Address - Likely C2 Endpoint

cltrixworkspace1[.]com - Domain - C2 Endpoint

vpnupdaters[.]com - Domain - C2 Endpoint

93.115.27[.]71 - IP Address - Possible Exfiltration Endpoint

158.51.85[.]157 - IP Address - Possible Exfiltration Endpoint

w.api.mega.co[.]nz - Domain - Possible Exfiltration Endpoint

*.userstorage.mega.co[.]nz - Domain - Possible Exfiltration Endpoint

741cc67d2e75b6048e96db9d9e2e78bb9a327e87 - SHA1 Hash - Hive Ransomware File

2f9da37641b204ef2645661df9f075005e2295a5 - SHA1 Hash - Likely Hive Ransomware File

hiveleakdbtnp76ulyhi52eag6c6tyc3xw7ez7iqy6wc34gd2nekazyd[.]onion - TOR Domain - Likely Hive Endpoint

References

[1] https://www.justice.gov/opa/pr/us-department-justice-disrupts-hive-ransomware-variant

[2] https://www.varonis.com/blog/hive-ransomware-analysis

[3] https://www.trendmicro.com/vinfo/us/security/news/ransomware-spotlight/ransomware-spotlight-hive 

[4]https://www.cisa.gov/news-events/cybersecurity-advisories/aa22-321a

[5] https://www.trendmicro.com/en_us/research/22/c/nokoyawa-ransomware-possibly-related-to-hive-.html

[6] https://www.virustotal.com/gui/file/60f6a63e366e6729e97949622abd9de6d7988bba66f85a4ac8a52f99d3cb4764/detection

[7] https://heimdalsecurity.com/blog/what-is-hive-ransomware/

[8] https://www.microsoft.com/en-us/security/blog/2022/07/05/hive-ransomware-gets-upgrades-in-rust/ 

Written by
Emily Megan Lim
Cyber 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
Emily Megan Lim
Cyber Analyst

Botnet Behind the Camera: Mirai Katana Activity on a Video Recording Device

OT
•
September 2, 2026
Parvatha Ananthakannan
Cyber Analyst
Watch the NIS2 Webinar
Trending blogs
1
Prompt Security in Enterprise AI: Strengths, Weaknesses, and Common Approaches
May 20, 2026
2
How Empowering End Users can Improve Your Email Security and Decrease the Burden on the SOC
May 8, 2024
3
Elevating Network Security: Confronting Trust, Ransomware, & Novel Attacks
Jun 21, 2024
4
The State of AI in Cybersecurity: Unveiling Global Insights from 1,800 Security Practitioners
Apr 9, 2024
5
The State of AI in Cybersecurity: The Impact of AI on Cybersecurity Solutions
May 13, 2024

More in this series

No items found.
Continue reading
Network
•
August 6, 2026

When AI Agents Attack: The Case for Behavioral Anomaly Detection

A new UK AI Security Institute report shows autonomous AI agents independently developing novel attack chains, adapting when detected, and targeting humans and other AI systems. As agents become embedded across the enterprise, behavioral anomaly detection will be critical to securing the new attack surface they create.
Adam Stevens
Senior Director of Product
Read more
Network
•
July 27, 2026

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

Darktrace detected a multi-stage ransomware intrusion from its earliest stages, identifying reconnaissance, privilege escalation, lateral movement, command-and-control communications, and data exfiltration before encryption occurred. This analysis highlights how behavioral detection exposed attacker activity using legitimate tools and infrastructure, providing multiple opportunities for early intervention and containment.
Alexandra Evzona
Cyber Analyst
Read more
Network
•
July 1, 2026

9 Stages of Ransomware & How AI Responds

Discover the 9 stages of ransomware attacks and how AI responds at each stage. Learn how you can protect your business from cyber threats.
The Darktrace Community
Read more

Blog

/

/

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

Blog

/

/

September 24, 2026

Agent Hijacks: Hijacking Agentic Harnesses to Attack an Organization

Default blog imageDefault blog image

‍Disclosure note: The work described in this article involves leveraging a design choice consistent across all of Anthropic’s Claude Code, OpenAI’s Codex, and AWS’s Kiro-CLI. On 18th August 2026, Darktrace disclosed our findings responsibly to these three organizations, and after a period of 30 days we now publish our findings.

Key takeaways:

  • Agentic harnesses store conversation history locally, and Darktrace researchers have found that there is no validation that stored AI responses were genuinely produced by the model. Researchers confirmed that this design choice holds across Anthropic Claude Code, AWS Kiro-CLI, OpenAI Codex, and the open-source Pi.
  • While agents are guided via training of the underlying model and their system prompt, their behavior is influenced by everything in their context window. Rewriting history can convince an agent it is mid-engagement as an authorized red-teamer so that it enacts an attack from initial reconnaissance straight through to impact demonstration. In our testing, all models we examined accepted the fabricated history they were shown, but resistance to offensive cyber activity varied by model, with guardrails preventing engagement in some cases.
  • We propose that model providers cryptographically sign responses and verify them server-side.  Since this fix is provider-side, defenders cannot deploy it themselves. Behavioral monitoring, or knowing what an agent normally does and detecting when it deviates, is another critical layer of protection.

Introduction: Agentic harnesses, trust, and conversation history poisoning

Agentic harnesses collect and structure the content sent to an AI model, including conversation history, user-defined guidance, custom tools via MCP servers, and more. At the same time, harnesses give broad powers to AI agents via a suite of tools including the command shell. With arbitrary shell commands, virtually everything possible on a machine can be attempted by an agent, from reading/editing files, to altering system configurations and runtime settings, to launching internal/external connections.

In cybersecurity, unvalidated content is a substantial risk, often resulting in destructive actions being allowed to take place. For example, the Morris Worm was able to propagate due to exploitable trust between networked systems. Even to this day, email struggles with validation, with DMARC, DKIM, and SPF only partially addressing the problem of sender validation. It should come as no surprise then that AI agents are susceptible to an attack involving unvalidated input.

Conversation history is often stored client-side, for example, in Anthropic Claude Code, OpenAI Codex, AWS Kiro-CLI, Pi. Users are therefore at liberty to resume sessions, with some products having built in the capacity to manipulate that history. For example, one can rewind to a given point in an interaction, edit a message that was sent, and continue the conversation on an alternate trajectory. Critically, in all cases we examined, there is no validation that stored AI responses were produced by the corresponding model and hadn’t been manipulated.  

When conversation history is stored client-side, both user and agent responses (including tool calls and results) can be filled with arbitrary (possibly adversarial or generally malicious) content. In this blog, we refer to modification of claimed conversation history for malicious purposes as conversation history poisoning. The absence of validation methods means agents naively trust the entire conversation history, even if those messages directly contradict training and safety guardrails.

Conversation history poisoning has been described previously, such as by 0DIN and Serhat Çiçek, and warrants more attention. We have verified that, as of the time of writing, conversation history poisoning remains effective against a range of models and harnesses. Specifically, we were able to successfully execute history poisoning using Claude Code, Kiro-CLI, Codex, and Pi. Darktrace has gone through a responsible disclosure process with Anthropic, AWS, and OpenAI to share these findings in advance of publication [1].

‍

Figure 1a: Left: the actual model response. Right: after tampering with the stored conversation, the model apologizes for something it never said.
Figure 1b: The conversation as stored in Kiro-CLI's SQLite database. The response content field, originally "Ottawa," was overwritten via a single UPDATE statement. The harness trusts the database without validation.

How we conducted the research

Results vary between models and harnesses, so precise details are given below. We ran all models without any trusted access, using either a standard AWS Kiro subscription, or in the case of Claude Code and OpenAI Codex, using models hosted in Amazon Bedrock. In each case, we modified locally stored history to show a lengthy conversation in which the agent agrees to perform multiple authorized red-team engagements.

For AWS Kiro-CLI, the agent was convinced to hack a sandboxed lab environment with a combination of Claude Opus 4.6 and Claude Sonnet 4.5. Ultimately, the full AD was compromised.

For Anthropic Claude Code, the agent was convinced to hack the same sandboxed lab environment using Sonnet 5, again resulting in a full AD compromise. Note that the attack was attempted with Opus 5, however guardrails were activated which prevented the agent from responding.

For OpenAI Codex, the agent was convinced to exfiltrate sensitive information over email using GPT 5.6 Sol. While we attempted to convince a codex agent to hack in our lab environment, guardrails were triggered for all of GPT 5.6 Luna, Terra, and Sol.

Agent Guardrails and Discretion

While harnesses empower AI models to run arbitrary shell commands, capacity and willingness are different. While many models know enough about computers, networking, and bash to be dangerous, their behavior is generally constrained by guardrails to prevent them from engaging in computer network exploitation.

Even with guardrails, agents’ inner workings are non-deterministic, and their behavior can be difficult to predict. Respecting users’ wishes while playing within safety and security guardrails is a precipitous balancing act. Many requests could be in service of either legitimate admin or malice. Asking an agent to reset a password is illustrative:  

‍

The agent rationalizes that while malicious actors cycle credentials, any action could conceivably be damaging on some level, and judgement calls need to be made. Ultimately, the agent agrees to reset the password. Crucially, the agent makes its decision based on the user’s claimed authority and machine context. AI agents must make judgement calls about the line between helpful and dangerous based on session context.

Agent hijack

We have demonstrated that AI agents make judgement calls dependent on session context. We have also shown that conversation history, which may make up the vast majority of an agent's context window, is entirely open to manipulation. Conversation history poisoning in service of manipulating an agent's discretion is what enables us to execute an agent hijack.  

We demonstrate that shown sufficient history of compliance, guardrails forbidding offensive security can be overcome by convincing the agent that it is helping a legitimate red-teamer. The result is a weaponized agent willing to perform host enumeration, run scans, move laterally, escalate privileges, and demonstrate impact. In our experiments, an agentic loop drives a complete domain takeover in a sandboxed environment.

‍

Left: the agent refuses when asked to perform network exploitation. Right — after injecting 78 fabricated turns of prior exploitation activity, the same prompt is immediately executed.

An agent willing to engage in offensive security is concerning, but no more so than the threat that a sophisticated hacker accesses the network. Consider, however, the following chain of events:

  1. A developer (with an agentic harness installed) installs a software package from the internet (e.g. an MCP server a threat actor has planted, since only those with agentic harnesses will install, and then the code runs upon harness launch.)
  2. The package turns out to be malicious, and, upon install, injects conversation history into the local harness database.
  3. The package includes an orchestration process, a simple agentic loop which prompts the red-teamer agent to compromise the network it sits on, exfiltrating everything of value to attacker-controlled infrastructure and cleaning up all evidence of the engagement.

Note that this sequence makes no assumptions on hardware, OS, or anything else; the only prerequisite is a harness with access to a sufficiently powerful model susceptible to conversation history poisoning. Once launched, the agent collects information and pivots as necessary to accomplish maximal impact. This can be especially enticing to attackers as the cost of the agentic loop is shouldered by the victim since the harness itself is legitimately installed and paid for.

Secure AI: Conversation history poisoning and beyond

Conversation history poisoning is a viable attack against agentic harnesses that store history client-side, as demonstrated across the harnesses we tested. Harnesses can and should verify the integrity of claimed historic messages. Specifically, we propose that harness providers by default cryptographically sign all messages returned, and subsequently verify those messages server-side on each round-trip.

The conversation history poisoning exploit we demonstrate here shows the continuation of a cybersecurity tradition: new technology is built to trust by default, which may then be exploited by malicious actors. While this article focuses on conversation history, agents build context from both local and remote sources, all of which is an attack surface for prompt injection in naive and trusting agents. Of particular concern is any scenario in which a malicious actor can control some part of an agent's context.

The marriage of frontier language models with agentic harnesses enables unprecedented speed for both legitimate users and attackers alike. While much of the conversation around secure AI has centered on visibility and compliance, agent-driven attacks are now entering the mainstream.

Darktrace / SECURE AI is our answer to this problem. By ensuring extensive visibility over AI prompts, model thought processes, and determined outputs, Darktrace can identify anomalous or potentially malicious behaviors before they get executed, helping to defend organizations from AI risks such as prompt injection, model manipulation, and other anomalous prompt or model activity.

‍

Footnotes

[1] We did not go through any responsible disclosure process with Pi. Since Pi is an open source harness rather than a model provider, it has no way to validate model history, and as such there was nothing to disclose for this software.

‍

[related-resource]

Continue reading
About the author
Eric Rozon
Senior Security Researcher
Your data. Our AI.
Elevate your network security with Darktrace AI