ブログ
/
Network
/
December 6, 2023

How Darktrace Triumphed Over MyKings Botnet

Darktrace has provided full visibility over the MyKings botnet kill chain from the beginning of its infections to the eventual cryptocurrency mining activity.
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
Oluwatosin Aturaka
Analyst Team Lead, Cambridge
Default blog image
06
Dec 2023

Botnets: A persistent cyber threat

Since their appearance in the wild over three decades ago, botnets have consistently been the attack vector of choice for many threat actors. The most prevalent of these attack vectors are distributed denial of service (DDoS) and phishing campaigns. Their persistent nature means that even if a compromised device in identified, attackers can continue to operate by using the additional compromised devices they will likely have on the target network. Similarly, command and control (C2) infrastructure can easily be restructured between infected systems, making it increasingly difficult to remove the infection.  

MyKings Botnet

One of the most prevalent and sophisticated examples in recent years is the MyKings botnet, also known as Smominru or DarkCloud. Darktrace has observed numerous cases of MyKings botnet compromises across multiple customer environments in several different industries as far back as August 2022. The diverse tactics, techniques, and procedures (TTPs) and sophisticated kill chains employed by MyKings botnet may prove a challenge to traditional rule and signature-based detections.

However, Darktrace’s anomaly-centric approach enabled it to successfully detect a wide-range of indicators of compromise (IoCs) related to the MyKings botnet and bring immediate awareness to customer security teams, as it demonstrated on the network of multiple customers between March and August 2023.

Background on MyKings Botnet

MyKings has been active and spreading steadily since 2016 resulting in over 520,000 infections worldwide.[1] Although verified attribution of the botnet remains elusive, the variety of targets and prevalence of crypto-mining software on affected devices suggests the threat group behind the malware is financially motivated. The operators behind MyKings appear to be highly opportunistic, with attacks lacking an obvious specific target industry. Across Darktrace’s customer base, the organizations affected were representative of multiple industries such as entertainment, mining, education, information technology, health, and transportation.

Given its longevity, the MyKings botnet has unsurprisingly evolved since its first appearance years ago. Initial analyses of the botnet showed that the primary crypto-related activity on infected devices was the installation of Monero-mining software. However, in 2019 researchers discovered a new module within the MyKings malware that enabled clipboard-jacking, whereby the malware replaces a user's copied cryptowallet address with the operator's own wallet address in order to siphon funds.[2]

Similar to other botnets such as the Outlaw crypto-miner, the MyKings botnet can also kill running processes of unrelated malware on the compromised hosts that may have resulted from prior infection.[3] MyKings has also developed a comprehensive set of persistence techniques, including: the deployment of bootkits, initiating the botnet immediately after a system reboot, configuring Registry run keys, and generating multiple Scheduled Tasks and WMI listeners.[4] MyKings have also been observed rotating tools and payloads over time to propagate the botnet. For example, some operators have been observed utilizing PCShare, an open-source remote access trojan (RAT) customized to conduct C2 services, execute commands, and download mining software[5].

Darktrace Coverage

Across observed customer networks between March and August 2023, Darktrace identified the MyKings botnet primarily targeting Windows-based servers that supports services like MySQL, MS-SQL, Telnet, SSH, IPC, WMI, and Remote Desktop (RDP).  In the initial phase of the attack, the botnet would initiate a variety of attacks against a target including brute-forcing and exploitation of unpatched vulnerabilities on exposed servers. The botnet delivers a variety of payloads to the compromised systems including worm downloaders, trojans, executable files and scripts.

This pattern of activity was detected across the network of one particular Darktrace customer in the education sector in early March 2023. Unfortunately, this customer did not have Darktrace RESPOND™ deployed on their network at the time of the attack, meaning the MyKings botnet was able to move through the cyber kill chain ultimately achieving its goal, which in this case was mining cryptocurrency.

Initial Access

On March 6, Darktrace observed an internet-facing SQL server receiving an unusually large number of incoming MySQL connections from the rare external endpoint 171.91.76[.]31 via port 1433. While it is not possible to confirm whether these suspicious connections represented the exact starting point of the infection, such a sudden influx of SQL connection from a rare external endpoint could be indicative of a malicious attempt to exploit vulnerabilities in the server's SQL database or perform password brute-forcing to gain unauthorized access. Given that MyKings typically spreads primarily through such targeting of internet-exposed devices, the pattern of activity is consistent with potential initial access by MyKings.[6]

Initial Command and Control

The device then proceeded to initiate a series of repeated HTTP connections between March 6 and March 10, to the domain www[.]back0314[.]ru (107.148.239[.]111). These connections included HTTP GET requests featuring URIs such as ‘/back.txt',  suggesting potential beaconing and C2 communication. The device continued this connectivity to the external host over the course of four days, primarily utilizing destination ports 80, and 6666. While port 80 is commonly utilized for HTTP connections, port 6666 is a non-standard port for the protocol. Such connectivity over non-standard ports can indicate potential detection evasion and obfuscation tactics by the threat actors.  During this time, the device also initiated repeated connections to additional malicious external endpoints with seemingly algorithmically generated hostnames such as pc.pc0416[.]xyz.

Darktrace UI image
Figure 1: Model breach showing details of the malicious domain generation algorithm (DGA) connections.

Tool Transfer

While this beaconing activity was taking place, the affected device also began to receive potential payloads from unusual external endpoints. On April 29, the device made an HTTP GET request for “/power.txt” to the endpoint 192.236.160[.]237, which was later discovered to have multiple open-source intelligence (OSINT) links to malware. Power.txt is a shellcode written in PowerShell which is downloaded and executed with the purpose of disabling Windows Defenders related functions.[7] After the initial script was downloaded (and likely executed), Darktrace went on to detect the device making a series of additional GET requests for several varying compressed and executable files. For example, the device made HTTP requests for '/pld/cmd.txt' to the external endpoint 104.233.224[.]173. In response the external server provided numerous files, including ‘u.exe’, and ‘upsup4.exe’ for download, both of which share file names with previously identified MyKings payloads.

MyKings deploys a diverse array of payloads to expand the botnet and secure a firm position within a compromised system. This multi-faceted approach may render conventional security measures less effective due to the intricacies of and variety of payloads involved in compromises. Darktrace, however, does not rely on static or outdated lists of IoCs in order to detect malicious activity. Instead, DETECT’s Self-Learning AI allows it to identify emerging compromise activity by recognizing the subtle deviations in an affected device’s behavior that could indicate it has fallen into the hands of malicious actors.

Figure 2: External site summary of the endpoint 103.145.106[.]242 showing the rarity of connectivity to the external host.

Achieving Objectives – Crypto-Mining

Several weeks after the initial payloads were delivered and beaconing commenced, Darktrace finally detected the initiation of crypto-mining operations. On May 27, the originally compromised server connected to the rare domain other.xmrpool[.]ru over port 1081. As seen in the domain name, this endpoint appears to be affiliated with pool mining activity and the domain has various OSINT affiliations with the cryptocurrency Monero coin. During this connection, the host was observed passing Monero credentials, activity which parallels similar mining operations observed on other customer networks that had been compromised by the MyKings botnet.

Although mining activity may not pose an immediate or urgent concern for security unauthorized cryptomining on devices can result in detrimental consequences, such as compromised hardware integrity, elevated energy costs, and reduced productivity, and even potential involvement in money laundering.

Figure 3: Event breach log showing details of the connection to the other.xmrpool[.]ru endpoint associated with cryptocurrency mining activity.

Conclusion

Detecting future iterations of the MyKings botnet will likely demand a shift away from an overreliance on traditional rules and signatures and lists of “known bads”, instead requiring organizations to employ AI-driven technology that can identify suspicious activity that represents a deviation from previously established patterns of life.

Despite the diverse range of payloads, malicious endpoints, and intricate activities that constitute a typical MyKing botnet compromise, Darktrace was able successfully detect multiple critical phases within the MyKings kill chain. Given the evolving nature of the MyKings botnet, it is highly probable the botnet will continue to expand and adapt, leveraging new tactics and technologies. By adopting Darktrace’s product of suites, including Darktrace DETECT, organizations are well-positioned to identify these evolving threats as soon as they emerge and, when coupled with the autonomous response technology of Darktrace RESPOND, threats like the MyKings botnet can be stopped in their tracks before they can achieve their ultimate goals.

Credit to: Oluwatosin Aturaka, Analyst Team Lead, Cambridge, Adam Potter, Cyber Analyst

Appendix

IoC Table

IoC - Type - Description + Confidence

162.216.150[.]108- IP - C2 Infrastructure

103.145.106[.]242 - IP - C2 Infrastructure

137.175.56[.]104 - IP - C2 Infrastructure

138.197.152[.]201 - IP - C2 Infrastructure

139.59.74[.]135 - IP - C2 Infrastructure

pc.pc0416[.]xyz - Domain - C2 Infrastructure (DGA)

other.xmrpool[.]ru - Domain - Cryptomining Endpoint

xmrpool[.]ru - Domain - Cryptomining Endpoint

103.145.106[.]55 - IP - Cryptomining Endpoint

ntuser[.]rar - Zipped File - Payload

/xmr1025[.]rar - Zipped File - Payload

/20201117[.]rar - Zipped File - Payload

wmi[.]txt - File - Payload

u[.]exe - Executable File - Payload

back[.]txt - File - Payload

upsupx2[.]exe - Executable File - Payload

cmd[.]txt - File - Payload

power[.]txt - File - Payload

ups[.]html - File - Payload

xmr1025.rar - Zipped File - Payload

171.91.76[.]31- IP - Possible Initial Compromise Endpoint

www[.]back0314[.]ru - Domain - Probable C2 Infrastructure

107.148.239[.]111 - IP - Probable C2 Infrastructure

194.67.71[.]99 - IP- Probable C2 Infrastructure

Darktrace DETECT Model Breaches

  • Device / Initial Breach Chain Compromise
  • Anomalous File / Masqueraded File Transfer (x37)
  • Compromise / Large DNS Volume for Suspicious Domain
  • Compromise / Fast Beaconing to DGA
  • Device / Large Number of Model Breaches
  • Anomalous File / Multiple EXE from Rare External Locations (x30)
  • Compromise / Beacon for 4 Days (x2)
  • Anomalous Server Activity / New User Agent from Internet Facing System
  • Anomalous Connection / New User Agent to IP Without Hostname
  • Anomalous Server Activity / New Internet Facing System
  • Anomalous File / EXE from Rare External Location (x37)
  • Device / Large Number of Connections to New Endpoints
  • Anomalous Server Activity / Server Activity on New Non-Standard Port (x3)
  • Device / Threat Indicator (x3)
  • Unusual Activity / Unusual External Activity
  • Compromise / Crypto Currency Mining Activity (x37)
  • Compliance / Internet Facing SQL Server
  • Device / Anomalous Scripts Download Followed By Additional Packages
  • Device / New User Agent

MITRE ATT&CK Mapping

ATT&CK Technique - Technique ID

Reconnaissance – T1595.002 Vulnerability Scanning

Resource Development – T1608 Stage Capabilities

Resource Development – T1588.001 Malware

Initial Access – T1190 Exploit Public-Facing Application

Command and Control – T15568.002 Domain Generated Algorithms

Command and Control – T1571 Non-Standard Port

Execution – T1047 Windows Management Instrumentation

Execution – T1059.001 Command and Scripting Interpreter

Persistence – T1542.003 Pre-OS Boot

Impact – T1496 Resource Hijacking

References

[1] https://www.binarydefense.com/resources/threat-watch/mykings-botnet-is-growing-and-remains-under-the-radar/

[2] https://therecord.media/a-malware-botnet-has-made-more-than-24-7-million-since-2019

[3] https://www.darktrace.com/blog/outlaw-returns-uncovering-returning-features-and-new-tactics

[4] https://www.sophos.com/en-us/medialibrary/pdfs/technical-papers/sophoslabs-uncut-mykings-report.pdf

[5] https://www.antiy.com/response/20190822.html

[6] https://ethicaldebuggers.com/mykings-botnet/

[7] https://ethicaldebuggers.com/mykings-botnet/

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
Oluwatosin Aturaka
Analyst Team Lead, Cambridge

More in this series

No items found.

Blog

/

AI

/

September 24, 2026

Detecting Rogue Agent Behavior in the Enterprise

Default blog imageDefault blog image

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

Key Takeaways:

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

Introduction: Understanding the Threat of Hacking by Agents

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

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

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

Setting the Stage

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

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

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

Inducing Cheating Behavior

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

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

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

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

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

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

Details from a Daybreak Test

‍

‍

Continue reading
About the author
Sam Lister
Specialist Security Researcher

Blog

/

AI

/

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
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