November 5, 2025

Tracking a Dragon: Investigating a DragonForce-affiliated ransomware attack with Darktrace

Darktrace investigates a DragonForce-affiliated ransomware attack targeting the manufacturing sector. The blog details the attack lifecycle, from network scanning and credential brute-forcing to data exfiltration and file encryption.

Written by

Justin Torres

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

Justin Torres

Cyber Analyst

Nov 2025

What is DragonForce?

DragonForce is a Ransomware-as-a-Service (RaaS) platform that emerged in late 2023, offering broad-scale capabilities and infrastructure to threat actors. Recently, DragonForce has been linked to attacks targeting the UK retail sector, resulting in several high-profile cases [1][2]. Moreover, the group launched an affiliate program offering a revenue share of roughly 20%, significantly lower than commissions reported across other RaaS platforms [3].

This Darktrace case study examines a DragonForce-linked RaaS infection within the manufacturing industry. The earliest signs of compromise were observed during working hours in August 2025, where an infected device started performing network scans and attempted to brute-force administrative credentials. After eight days of inactivity, threat actors returned and multiple devices began encrypting files via the SMB protocol using a DragonForce-associated file extension. Ransom notes referencing the group were also dropped, suggesting the threat actor is claiming affiliation with DragonForce, though this has not been confirmed.

Despite Darktrace’s detection of the attack in its early stages, the customer’s deployment did not have Darktrace’s Autonomous Response capability configured, allowing the threat to progress to data exfiltration and file encryption.

Darktrace's Observations

While the initial access vector was not clearly defined in this case study, it was likely achieved through common methods previously employed out by DragonForce affiliates. These include phishing emails leveraging social engineering tactics, exploitation of public-facing applications with known vulnerabilities, web shells, and/or the abuse of remote management tools.

Darktrace’s analysis identified internal devices performing internal network scanning, brute-forcing credentials, and executing unusual Windows Registry operations. Notably, Windows Registry events involving "Schedule\Taskcache\Tasks" contain subkeys for individual tasks, storing GUIDs that can be used to locate and analyze scheduled tasks. Additionally, Control\WMI\Security holds security descriptors for WMI providers and Event Tracing loggers that use non-default security settings respectively.

Furthermore, Darktrace identified data exfiltration activity over SSH, including connections to an ASN associated with a malicious hosting service geolocated in Russia.

1. Network Scan & Brute Force

Darktrace identified anomalous behavior in late August to early September 2025, originating from a source device engaging in internal network scanning followed by brute-force attempts targeting administrator credential, including “administrator”, “Admin”, “rdpadmin”, “ftpadmin”.

Upon further analysis, one of the HTTP connections seen in this activity revealed the use of the user agent string “OpenVAS-VT”, suggesting that the device was using the OpenVAS vulnerability scanner. Subsequently, additional devices began exhibiting network scanning behavior. During this phase, a file named “delete.me” was deleted by multiple devices using SMB protocol. This file is commonly associated with network scanning and penetration testing tool NetScan.

2. Windows Registry Key Update

Following the scanning phase, Darktrace observed the initial device then performing suspicious Winreg operations. This included the use of the ”BaseRegOpenKey” function across multiple registry paths.

Additional operations such as “BaseRegOpenKey” and “BaseRegQueryValue” were also seen around this time. These operations are typically used to retrieve specific registry key values and allow write operations to registry keys.

The registry keys observed included “SYSTEM\CurrentControlSet\Control\WMI\Security” and “Software\Microsoft\Windows NT\CurrentVersion\Schedule\Taskcache\Tasks”. These keys can be leveraged by malicious actors to update WMI access controls and schedule malicious tasks, respectively, both of which are common techniques for establishing persistence within a compromised system.

3. New Administrator Credential Usage

Darktrace subsequently detected the device using a highly privileged credential, “administrator”, via a successful Kerberos login for the first time. Shortly after, the same credential was used again for a successful SMB session.

These marked the first instances of authentication using the “administrator” credential across the customer’s environment, suggesting potential malicious use of the credential following the earlier brute-force activity.

Figure 1: Darktrace’s detection of administrator credentials being used in Kerberos login events by an infected device.

Figure 2: Darktrace’s detection of administrator credentials being used in SMB sessions by an infected device.

4. Data Exfiltration

Prior to ransomware deployment, several infected devices were observed exfiltrating data to the malicious IP 45.135.232[.]229 via SSH connections [7][8]. This was followed by the device downloading data from other internal devices and transferring an unusually large volume of data to the same external endpoint.

The IP address was first seen on the network on September 2, 2025 - the same date as the observed data exfiltration activity preceding ransomware deployment and encryption.

Further analysis revealed that the endpoint was geolocated in Russia and registered to the malicious hosting provider Proton66. Multiple external researchers have reported malicious activity involving the same Proton66 ASN (AS198953 Proton66 OOO) as far back as April 2025. These activities notably included vulnerability scanning, exploitation attempts, and phishing campaigns, which ultimately led to malware [4][5][6].

Data Exfiltration Endpoint details.

Endpoint: 45.135.232[.]229
ASN: AS198953 Proton66 OOO
Transport protocol: TCP
Application protocol: SSH
Destination port: 22

Figure 3: Darktrace’s summary of the external IP 45.135.232[.]229, first detected on September 2, 2025. The right-hand side showcases model alerts triggered related to this endpoint including multiple data exfiltration related model alerts.

Further investigation into the endpoint using open-source intelligence (OSINT) revealed that it led to a Microsoft Internet Information Services (IIS) Manager console webpage. This interface is typically used to configure and manage web servers. However, threat actors have been known to exploit similar setups, using fake certificate warnings to trick users into downloading malware, or deploying malicious IIS modules to steal credentials.

Figure 4: Live screenshot of the destination (45.135.232[.]229), captured via OSINT sources, displaying a Microsoft IIS Manager console webpage.

5. Ransomware Encryption & Ransom Note

Multiple devices were later observed connecting to internal devices via SMB and performing a range of actions indicative of file encryption. This suspicious activity prompted Darktrace’s Cyber AI Analyst to launch an autonomous investigation, during which it pieced together associated activity and provided concrete timestamps of events for the customer’s visibility.

During this activity, several devices were seen writing a file named “readme.txt” to multiple locations, including network-accessible webroot paths such as inetpub\ and wwwroot\. This “readme.txt” file, later confirmed to be the ransom note, claimed the threat actors were affiliated with DragonForce.

At the same time, devices were seen performing SMB Move, Write and ReadWrite actions involving files with the “.df_win” extension across other internal devices, suggesting that file encryption was actively occurring.

Figure 5: Darktrace’s detection of SMB events (excluding Read events) where the device was seen moving or writing files with the “.df_win” extension.

Figure 6: Darktrace’s detection of a spike in SMB Write events with the filename “readme.txt” on September 9, indicating the start of file encryption.

Conclusion

The rise of Ransomware-as-a-Service (RaaS) and increased attacker customization is fragmenting tactics, techniques, and procedures (TTPs), making it increasingly difficult for security teams to prepare for and defend against each unique intrusion. RaaS providers like DragonForce further complicate this challenge by enabling a wide range of affiliates, each with varying levels of sophistication [9].

In this instance, Darktrace was able to identify several stages of the attack kill chain, including network scanning, the first-time use of privileged credentials, data exfiltration, and ultimately ransomware encryption. Had the customer enabled Darktrace’s Autonomous Response capability, it would have taken timely action to interrupt the attack in its early stages, preventing the eventual data exfiltration and ransomware detonation.

Credit to Justin Torres, Senior Cyber Analyst, Nathaniel Jones, VP, Security & AI Strategy, FCISO, & Emma Foulger, Global Threat Research Operations Lead.

Edited by Ryan Traill (Analyst Content Lead)

[related-resource]

Appendices

References:

1. https://www.infosecurity-magazine.com/news/dragonforce-goup-ms-coop-harrods/

2. https://www.picussecurity.com/resource/blog/dragonforce-ransomware-attacks-retail-giants

3. https://blog.checkpoint.com/security/dragonforce-ransomware-redefining-hybrid-extortion-in-2025/

4. https://www.trustwave.com/en-us/resources/blogs/spiderlabs-blog/proton66-part-1-mass-scanning-and-exploit-campaigns/

5. https://www.trustwave.com/en-us/resources/blogs/spiderlabs-blog/proton66-part-2-compromised-wordpress-pages-and-malware-campaigns/

6. https://www.broadcom.com/support/security-center/protection-bulletin/proton66-infrastructure-tied-to-expanding-malware-campaigns-and-c2-operations

7. https://www.virustotal.com/gui/ip-address/45.135.232.229

8. https://spur.us/context/45.135.232.229

9. https://www.group-ib.com/blog/dragonforce-ransomware/

IoC - Type - Description + Confidence

· 45.135.232[.]229 - Endpoint Associated with Data Exfiltration

· .readme.txt – Ransom Note File Extension

· .df_win – File Encryption Extension Observed

MITRE ATT&CK Mapping

DragonForce TTPs vs Darktrace Models

Initial Access:

· Anomalous Connection::Callback on Web Facing Device

Command and Control:

· Compromise::SSL or HTTP Beacon

· Compromise::Beacon to Young Endpoint

· Compromise::Beaconing on Uncommon Port

· Compromise::Suspicious SSL Activity

· Anomalous Connection::Devices Beaconing to New Rare IP

· Compromise::Suspicious HTTP and Anomalous Activity

· DNS Tunnel with TXT Records

Tooling:

· Anomalous File::EXE from Rare External Location

· Anomalous File::Masqueraded File Transfer

· Anomalous File::Numeric File Download

· Anomalous File::Script from Rare External Location

· Anomalous File::Uncommon Microsoft File then Exe

· Anomalous File::Zip or Gzip from Rare External Location

· Anomalous File::Uncommon Microsoft File then Exe

· Anomalous File::Internet Facing System File Download

Reconnaissance:

· Device::Suspicious SMB Query

· Device::ICMP Address Scan

· Anomalous Connection::SMB Enumeration

· Device::Possible SMB/NTLM Reconnaissance

· Anomalous Connection::Possible Share Enumeration Activity

· Device::Possible Active Directory Enumeration

· Anomalous Connection::Large Volume of LDAP Download

· Device::Suspicious LDAP Search Operation

Lateral Movement:

· User::Suspicious Admin SMB Session

· Anomalous Connection::Unusual Internal Remote Desktop

· Anomalous Connection::Unusual Long Remote Desktop Session

· Anomalous Connection::Unusual Admin RDP Session

· User::New Admin Credentials on Client

· User::New Admin Credentials on Server

· Multiple Device Correlations::Spreading New Admin Credentials

· Anomalous Connection::Powershell to Rare External

· Device::New PowerShell User Agent

· Anomalous Active Directory Web Services

· Compromise::Unusual SVCCTL Activity

Evasion:

· Unusual Activity::Anomalous SMB Delete Volume

· Persistence

· Device::Anomalous ITaskScheduler Activity

· Device::AT Service Scheduled Task

· Actions on Objectives

· Compromise::Ransomware::Suspicious SMB Activity (EM)

· Anomalous Connection::Sustained MIME Type Conversion

· Compromise::Ransomware::SMB Reads then Writes with Additional Extensions

· Compromise::Ransomware::Possible Ransom Note Write

· Data Sent to Rare Domain

· Uncommon 1 GiB Outbound

· Enhanced Unusual External Data Transfer

Darktrace Cyber AI Analyst Coverage/Investigation Events:

· Web Application Vulnerability Scanning of Multiple Devices

· Port Scanning

· Large Volume of SMB Login Failures

· Unusual RDP Connections

· Widespread Web Application Vulnerability Scanning

· Unusual SSH Connections

· Unusual Repeated Connections

· Possible Application Layer Reconnaissance Activity

· Unusual Administrative Connections

· Suspicious Remote WMI Activity

· Extensive Unusual Administrative Connections

· Suspicious Directory Replication Service Activity

· Scanning of Multiple Devices

· Unusual External Data Transfer

· SMB Write of Suspicious File

· Suspicious Remote Service Control Activity

· Access of Probable Unencrypted Password Files

· Internal Download and External Upload

· Possible Encryption of Files over SMB

· SMB Writes of Suspicious Files to Multiple Devices

_{The content provided in this blog is published by Darktrace for general informational purposes only and reflects our understanding of cybersecurity topics, trends, incidents, and developments at the time of publication. While we strive to ensure accuracy and relevance, the information is provided “as is” without any representations or warranties, express or implied. Darktrace makes no guarantees regarding the completeness, accuracy, reliability, or timeliness of any information presented and expressly disclaims all warranties.}

_{Nothing in this blog constitutes legal, technical, or professional advice, and readers should consult qualified professionals before acting on any information contained herein. Any references to third-party organizations, technologies, threat actors, or incidents are for informational purposes only and do not imply affiliation, endorsement, or recommendation.}

_{Darktrace, its affiliates, employees, or agents shall not be held liable for any loss, damage, or harm arising from the use of or reliance on the information in this blog.}

_{The cybersecurity landscape evolves rapidly, and blog content may become outdated or superseded. We reserve the right to update, modify, or remove any content.}

Get the latest insights on emerging cyber threats

This report explores the latest trends shaping the cybersecurity landscape and what defenders need to know in 2026.

Inside the SOC

Written by

Justin Torres

Cyber Analyst

•

September 2, 2026

Parvatha Ananthakannan

Specialist Security Researcher

Watch the NIS2 Webinar

Blog

AI

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

Blog

AI

September 24, 2026

Agent Hijacks: Hijacking Agentic Harnesses to Attack an Organization

‍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 18^th 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:

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.)
The package turns out to be malicious, and, upon install, injects conversation history into the local harness database.
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]

About the author

Eric Rozon

Senior 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

デモを予約

Check out this article by Darktrace: Tracking a Dragon: Investigating a DragonForce-affiliated ransomware attack with Darktrace

Tracking a Dragon: Investigating a DragonForce-affiliated ransomware attack with Darktrace

What is DragonForce?

Darktrace's Observations

1. Network Scan & Brute Force

2. Windows Registry Key Update

3. New Administrator Credential Usage

4. Data Exfiltration

5. Ransomware Encryption & Ransom Note

Conclusion

Appendices

References:

IoC - Type - Description + Confidence

MITRE ATT&CK Mapping

Get the latest insights on emerging cyber threats

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

Detecting Rogue Agent Behavior in the Enterprise

ブログをお楽しみ頂けましたか？

More in this series

When AI Agents Attack: The Case for Behavioral Anomaly Detection

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

9 Stages of Ransomware & How AI Responds

Blog

AI

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

Blog

AI

September 24, 2026

Agent Hijacks: Hijacking Agentic Harnesses to Attack an Organization

Key takeaways:

Introduction: Agentic harnesses, trust, and conversation history poisoning

How we conducted the research

Agent Guardrails and Discretion

Agent hijack

Secure AI: Conversation history poisoning and beyond

Footnotes