October 3, 2023

Unveiling ViperSoftX: A Darktrace Investigation

Read about the ViperSoftX threat and how Darktrace's innovative detection methods exposed this cyber intrusion and its potential impacts.

Written by

Zoe Tilsiter

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

Zoe Tilsiter

Cyber Analyst

Oct 2023

Fighting Info-Stealing Malware

The escalating threat posed by information-stealing malware designed to harvest and steal the sensitive data of individuals and organizations alike has become a paramount concern for security teams across the threat landscape. In direct response to security teams improving their threat detection and prevention capabilities, threat actors are forced to continually adapt and advance their techniques, striving for greater sophistication to ensure they can achieve the malicious goals.

What is ViperSoftX?

ViperSoftX is an information stealer and Remote Access Trojan (RAT) malware known to steal privileged information such as cryptocurrency wallet addresses and password information stored in browsers and password managers. It is commonly distributed via the download of cracked software from multiple sources such as suspicious domains, torrent downloads, and key generators (keygens) from third-party sites.

ViperSoftX was first observed in the wild in 2020 [1] but more recently, new strains were identified in 2022 and 2023 utilizing more sophisticated detection evasion techniques, making it more difficult for security teams to identify and analyze. This includes using more advanced encryption methods alongside monthly changes to command-and-control servers (C2) [2], using dynamic-link library (DLL) sideloading for execution techiques, and subsequently loading a malicious browser extension upon infection which works as an independent info-stealer named VenomSoftX [3].

Between February and June 2023, Darktrace detected activity related to the VipersoftX and VenomSoftX information stealers on the networks of more than 100 customers across its fleet. Darktrace DETECT™ was able to successfully identify the anomalous network activity surrounding these emerging information stealer infections and bring them to the attention of the customers, while Darktrace RESPOND™, when enabled in autonomous response mode, was able to quickly intervene and shut down malicious downloads and data exfiltration attempts.

ViperSoftX Attack & Darktrace Coverage

In cases of ViperSoftX information stealer activity observed by Darktrace, the initial infection was caused through the download of malicious files from multimedia sites, endpoints of cracked software like Adobe Illustrator, and torrent sites. Endpoint users typically unknowingly download the malware from these endpoints with a sideloaded DLL, posing as legitimate software executables.

Darktrace detected multiple downloads from such multimedia sites and endpoints related to cracked software and BitTorrent, which were likely representative of the initial source of ViperSoftX infection. Darktrace DETECT models such as ‘Anomalous File / Anomalous Octet Stream (No User Agent)’ breached in response to this activity and were brought to the immediate attention of customer security teams. In instances where Darktrace RESPOND was configured in autonomous response mode, Darktrace was able to enforce a pattern of life on offending devices, preventing them from downloading malicious files. This ensures that devices are limited to conducting only their pre-established expected activit, minimizing disruption to the business whilst targetedly mitigating suspicious file downloads.

The downloads are then extracted, decrypted and begin to run on the device. The now compromised device will then proceed to make external connections to C2 servers to retrieve secondary PowerShell executable. Darktrace identified that infected devices using PowerShell user agents whilst making HTTP GET requests to domain generation algorithm (DGA) ViperSoftX domains represented new, and therefore unusual, activity in a large number of cases.

For example, Darktrace detected one customer device making an HTTP GET request to the endpoint ‘chatgigi2[.]com’, using the PowerShell user agent ‘Mozilla/5.0 (Windows NT; Windows NT 10.0; en-US) WindowsPowerShell/5.1.19041.2364’. This new activity triggered a number of DETECT models, including ‘Anomalous Connection / PowerShell to Rare External’ and ‘Device / New PowerShell User Agent’. Repeated connections to these endpoints also triggered C2 beaconing models including:

Compromise / Agent Beacon (Short Period)
Compromise / Agent Beacon (Medium Period)
Compromise / Agent Beacon (Long Period)
Compromise / Quick and Regular Windows HTTP Beaconing
Compromise / SSL or HTTP Beacon

Although a large number of different DGA domains were detected, commonalities in URI formats were seen across affected customers which matched formats previously identified as ViperSoftX C2 communication by open-source intelligence (OSINT), and in other Darktrace investigations.

URI paths for example, were always of the format /api/, /api/v1/, /v2/, or /v3/, appearing to detail version number, as can be seen in Figure 1.

*Figure 1: A Packet Capture (PCAP) taken from Darktrace showing a connection made to a ViperSoftX C2 endpoint containing versioning information, consistent with ViperSoftX pattern of communication.*

Before the secondary PowerShell executables are loaded, ViperSoftX takes a digital fingerprint of the infected machine to gather its configuration details, and exfiltrates them to the C2 server. These include the computer name, username, Operating System (OS), and ensures there are no anti-virus or montoring tools on the device. If no security tool are detected, ViperSoftX then downloads, decrypts and executes the PowerShell file.

Following the GET requests Darktrace observed numerous devices performing HTTP POST requests and beaconing connections to ViperSoftX endpoints with varying globally unique identifiers (GUIDs) within the URIs. These connections represented the exfiltration of device configuration details, such as “anti-virus detected”, “app used”, and “device name”. As seen on another customer’s deployment, this caused the model ‘Anomalous Connection / Multiple HTTP POSTs to Rare Hostname’ to breach, which was also detected by Cyber AI Analyst as seen in Figure 2.

*Figure 2: Cyber AI Analyst’s detection of HTTP POSTs being made to apibiling[.]com, a ViperSoftX C2 endpoint.*

The malicious PowerShell download then crawls the infected device’s systems and directories looking for any cryptocurrency wallet information and password managers, and exfiltrates harvest data to the C2 infrastructure. The C2 server then provides further browser extensions to Chromium browsers to be downloaded and act as a separate stand-alone information stealer, also known as VenomSoftX.

Similar to the initial download of ViperSoftX, these malicious extensions are disguised as legitimate browser extensions to evade the detection of security teams. VenomSoft X, in turn, searches through and attempts to gather sensitive data from password managers and crypto wallets stored in user browsers. Using this information, VenomSoftX is able to redirect crypocurrency transactions by intercepting and manipulating API requests between the sender and the intended recipient, directing the cryptocurrency to the attacker instead [3].

Following investigation into VipersoftX activity across the customer base, Darktrace notified all affected customers and opened Ask the Expert (ATE) tickets through which customer’s could directly contact the analyst team for support and guidance in the face on the information stealer infection.

How did the attack bypass the rest of the security stack?

As previously mentioned, both the initial download of ViperSoftX and the subsequent download of the VenomX browser extension are disguised as legitimate software or browser downloads. This is a common technique employed by threat actors to infect target devices with malicious software, while going unnoticed by security teams traditional security measures. Furthermore, by masquerading as a legitimate piece of software endpoint users are more likely to trust and therefore download the malware, increasing the likelihood of threat actor’s successfully carrying out their objectives. Additionally, post-infection analysis of shellcode, the executable code used as the payload, is made significantly more difficult by VenomSoftX’s use of bytemapping. Bytemapping prevents the encryption of shellcodes without its corresponding byte map, meaning that the payloads cannot easily be decrypted and analysed by security researchers. [3]

ViperSoftX also takes numerous attempts to prevent their C2 infrastructure from being identified by blocking access to it on browsers, and using multiple DGA domains, thus renderring defunct traditional security measures that rely on threat intelligence and static lists of indicators of compromise (IoCs).

Fortunately for Darktrace customers, Darktrace’s anomaly-based approach to threat detection means that it was able to detect and alert customers to this suspicious activity that may have gone unnoticed by other security tools.

Insights/Conclusion

Faced with the challenge of increasingly competent and capable security teams, malicious actors are having to adopt more sophisticated techniques to successfully compromise target systems and achieve their nefarious goals.

ViperSoftX information stealer makes use of numerous tactics, techniques and procedures (TTPs) designed to fly under the radar and carry out their objectives without being detected. ViperSoftX does not rely on just one information stealing malware, but two with the subsequent injection of the VenomSoftX browser extension, adding an additional layer of sophistication to the informational stealing operation and increasing the potential yield of sensitive data. Furthermore, the use of evasion techniques like disguising malicious file downloads as legitimate software and frequently changing DGA domains means that ViperSoftX is well equipped to infiltrate target systems and exfiltrate confidential information without being detected.

However, the anomaly-based detection capabilities of Darktrace DETECT allows it to identify subtle changes in a device’s behavior, that could be indicative of an emerging compromise, and bring it to the customer’s security team. Darktrace RESPOND is then autonomously able to take action against suspicious activity and shut it down without latency, minimizing disruption to the business and preventing potentially significant financial losses.

Credit to: Zoe Tilsiter, Senior Cyber Analyst, Nathan Lorenzo, Cyber Analyst.

Appendices

References

[1] https://www.fortinet.com/blog/threat-research/vipersoftx-new-javascript-threat

[2] https://www.trendmicro.com/en_us/research/23/d/vipersoftx-updates-encryption-steals-data.html

[3] https://decoded.avast.io/janrubin/vipersoftx-hiding-in-system-logs-and-spreading-venomsoftx/

Darktrace DETECT Model Detections

· Anomalous File / Anomalous Octet Stream (No User Agent)

· Anomalous Connection / PowerShell to Rare External

· Anomalous Connection / Multiple HTTP POSTs to Rare Hostname

· Anomalous Connection / Lots of New Connections

· Anomalous Connection / Multiple Failed Connections to Rare Endpoint

· Anomalous Server Activity / Outgoing from Server

· Compromise / Large DNS Volume for Suspicious Domain

· Compromise / Quick and Regular Windows HTTP Beaconing

· Compromise / Beacon for 4 Days

· Compromise / Suspicious Beaconing Behaviour

· Compromise / Large Number of Suspicious Failed Connections

· Compromise / Large Number of Suspicious Successful Connections

· Compromise / POST and Beacon to Rare External

· Compromise / DGA Beacon

· Compromise / Agent Beacon (Long Period)

· Compromise / Agent Beacon (Medium Period)

· Compromise / Agent Beacon (Short Period)

· Compromise / Fast Beaconing to DGA

· Compromise / SSL or HTTP Beacon

· Compromise / Slow Beaconing Activity To External Rare

· Compromise / Beaconing Activity To External Rare

· Compromise / Excessive Posts to Root

· Compromise / Connections with Suspicious DNS

· Compromise / HTTP Beaconing to Rare Destination

· Compromise / High Volume of Connections with Beacon Score

· Compromise / Sustained SSL or HTTP Increase

· Device / New PowerShell User Agent

· Device / New User Agent and New IP

Darktrace RESPOND Model Detections

· Antigena / Network / External Threat / Antigena Suspicious File Block

· Antigena / Network / External Threat / Antigena File then New Outbound Block

· Antigena / Network / External Threat / Antigena Watched Domain Block

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

· Antigena / Network / External Threat / Antigena Suspicious Activity Block

· Antigena / Network / Significant Anomaly / Antigena Breaches Over Time Block

· Antigena / Network / Insider Threat / Antigena Large Data Volume Outbound Block

· Antigena / Network / External Threat / Antigena Suspicious File Pattern of Life Block

· Antigena / Network / Significant Anomaly / Antigena Controlled and Model Breach

List of IoCs

Indicator - Type - Description

ahoravideo-blog[.]com - Hostname - ViperSoftX C2 endpoint

ahoravideo-blog[.]xyz - Hostname - ViperSoftX C2 endpoint

ahoravideo-cdn[.]com - Hostname - ViperSoftX C2 endpoint

ahoravideo-cdn[.]xyz - Hostname - ViperSoftX C2 endpoint

ahoravideo-chat[.]com - Hostname - ViperSoftX C2 endpoint

ahoravideo-chat[.]xyz - Hostname - ViperSoftX C2 endpoint

ahoravideo-endpoint[.]xyz - Hostname - ViperSoftX C2 endpoint

ahoravideo-schnellvpn[.]com - Hostname - ViperSoftX C2 endpoint

ahoravideo-schnellvpn[.]xyz - Hostname - ViperSoftX C2 endpoint

apibilng[.]com - Hostname - ViperSoftX C2 endpoint

arrowlchat[.]com - Hostname - ViperSoftX C2 endpoint

bideo-blog[.]com - Hostname - ViperSoftX C2 endpoint

bideo-blog[.]xyz - Hostname - ViperSoftX C2 endpoint

bideo-cdn[.]com - Hostname - ViperSoftX C2 endpoint

bideo-cdn[.]xyz - Hostname - ViperSoftX C2 endpoint

bideo-chat[.]com - Hostname - ViperSoftX C2 endpoint

bideo-chat[.]xyz - Hostname - ViperSoftX C2 endpoint

bideo-endpoint[.]com - Hostname - ViperSoftX C2 endpoint

bideo-endpoint[.]xyz - Hostname - ViperSoftX C2 endpoint

bideo-schnellvpn[.]com - Hostname - ViperSoftX C2 endpoint

chatgigi2[.]com - Hostname - ViperSoftX C2 endpoint

counter[.]wmail-service[.]com - Hostname - ViperSoftX C2 endpoint

fairu-cdn[.]xyz - Hostname - ViperSoftX C2 endpoint

fairu-chat[.]xyz - Hostname - ViperSoftX C2 endpoint

fairu-endpoint[.]com - Hostname - ViperSoftX C2 endpoint

fairu-schnellvpn[.]com - Hostname - ViperSoftX C2 endpoint

fairu-schnellvpn[.]xyz - Hostname - ViperSoftX C2 endpoint

privatproxy-blog[.]com - Hostname - ViperSoftX C2 endpoint

privatproxy-blog[.]xyz - Hostname - ViperSoftX C2 endpoint

privatproxy-cdn[.]com - Hostname - ViperSoftX C2 endpoint

privatproxy-cdn[.]xyz - Hostname - ViperSoftX C2 endpoint

privatproxy-endpoint[.]xyz - Hostname - ViperSoftX C2 endpoint

privatproxy-schnellvpn[.]com - Hostname - ViperSoftX C2 endpoint

privatproxy-schnellvpn[.]xyz - Hostname - ViperSoftX C2 endpoint

static-cdn-349[.]net - Hostname - ViperSoftX C2 endpoint

wmail-blog[.]com - Hostname - ViperSoftX C2 endpoint

wmail-cdn[.]xyz - Hostname - ViperSoftX C2 endpoint

wmail-chat[.]com - Hostname - ViperSoftX C2 endpoint

wmail-schnellvpn[.]com - Hostname - ViperSoftX C2 endpoint

wmail-schnellvpn[.]xyz - Hostname - ViperSoftX C2 endpoint

Mozilla/5.0 (Windows NT; Windows NT 10.0; en-US) WindowsPowerShell/5.1.19041.2364 - User Agent -PowerShell User Agent

MITRE ATT&CK Mapping

‍Tactic - Technique - Notes

Command and Control - T1568.002 Dynamic Resolution: Domain Generation Algorithms

Command and Control - T1321 Data Encoding

Credential Access - T1555.005 Credentials from Password Stores: Password Managers

Defense Evasion - T1027 Obfuscated Files or Information

Execution - T1059.001 Command and Scripting Interpreter: PowerShell

Execution - T1204 User Execution T1204.002 Malicious File

Persistence - T1176 Browser Extensions - VenomSoftX specific

Persistence, Privilege Escalation, Defense Evasion - T1574.002 Hijack Execution Flow: DLL Side-Loading

Written by

Zoe Tilsiter

Cyber Analyst

Inside the SOC

Written by

Zoe Tilsiter

Cyber Analyst

•

September 2, 2026

Parvatha Ananthakannan

Specialist Security Researcher

Watch the NIS2 Webinar

Blog

September 24, 2026

Detecting Rogue Agent Behavior in the Enterprise

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

Key Takeaways:

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

Introduction: Understanding the Threat of Hacking by Agents

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

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

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

Setting the Stage

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

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

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

Inducing Cheating Behavior

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

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

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

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

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

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

Details from a Daybreak Test

‍

About the author

Sam Lister

Specialist Security Researcher

Blog

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

Your data. Our AI.

Elevate your network security with Darktrace AI

Get a demo

Check out this article by Darktrace: Unveiling ViperSoftX: A Darktrace Investigation

Unveiling ViperSoftX: A Darktrace Investigation

Fighting Info-Stealing Malware

What is ViperSoftX?

ViperSoftX Attack & Darktrace Coverage

How did the attack bypass the rest of the security stack?

Insights/Conclusion

Appendices

References

Darktrace DETECT Model Detections

Darktrace RESPOND Model Detections

List of IoCs

MITRE ATT&CK Mapping

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

Detecting Rogue Agent Behavior in the Enterprise

Enjoying the blog?

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

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

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