Blog
/
Email
/
August 7, 2024

How Darktrace’s AI Applies a Zero-Trust Mentality within Critical Infrastructure Supply Chains

Darktrace prevented a Critical National Infrastructure organization from falling victim to a SharePoint phishing attack originating from one of its trusted suppliers. This blog discusses common perceptions of zero-trust in email security, how AI that uses anomaly-based threat detection embodies core zero-trust principles and the relevance of this approach to securing CNI bodies with complex but interdependent supply chains from Cloud account compromise. 
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
Nicole Wong
Cyber Security Analyst
Default blog image
07
Aug 2024

Note: In order to name anonymity, real organization names have been replaced, all names used in this blog are fictitious.

What are critical national infrastructure sectors?

Critical National Infrastructure (CNI) sectors encompass of assets, systems, and networks essential to the functioning of society. Any disruption or destruction of these sectors could have wide-reaching and potentially disastrous effects on a country’s economy, security and/or healthcare services [1].

Cyber risks across Transportation Systems sector

Transportation Systems is one such CNI sector comprising of interconnected networks of fixed and mobile assets managed by both public and private operators. These systems are highly interdependent with other CNI sectors too. As such, the digital technologies this sector relies on – such as positioning and tracking, signaling, communications, industrial system controls, and data and business management – are often interconnected through different networks and remote access terminals. This interconnectedness creates multiple entry points that need to be security across the supply.

Digital transformation has swept through CNI sectors in recent years, including Transportation Systems. These organizations are now increasingly dependent on third-party and cloud providers for data storage and transmission, making their supply chains vulnerable to exploitation by malicious actors [2].

The exploitation of legitimate and popular cloud services mirrors the well-known “living-off-the-land” techniques, which are not being adapted to the cloud along with the resources they support. In one recent case previously discussed by Darktrace, for example, a phishing attack attempted to abuse Dropbox to deliver malicious payloads.

Zero-Trust within CNI Sectors

One recommended approach to secure an organization’s supply chain and cloud environments is the implementation of zero-trust strategies, which remove inherent trust within the network [3] [4]. The principle of “never trust, always verify” is widely recognized as an architectural design, with 63% of organizations surveyed by Gartner reportedly implementing a zero-trust strategy, but in most cases to less than 50% of their environments [5]

Although this figure reflects the reality and challenge of balancing operations and security, demands from the threat landscape and supply chain risks mean that organizations must adopt zero-trust principles in areas not traditionally considered part of network architecture, such as email and cloud environments.

Email is often the primary entry point for cyber-attacks with Business Email Compromise (BEC) being a major threat to CNI organizations. However, the application of zero-trust principles to secure email environments is still not well understood. Common misconceptions include:

  • “Positively identifying known and trusted senders” – Maintaining a list of “known and trusted senders” contradicts the zero-trust model, which assumes that no entity is inherently trustworthy.
  • “Using DMARC, DKIM and SPF” – While these protocols offer some protection, they are often insufficient on their own, as they can be bypassed and do not protect against email account takeovers. Research published from Darktrace’s last two threat reports consistently shows that at least 60% of phishing emails detected by Darktrace had bypassed Domain-based Message Authentication, Reporting & Conformance (DMARC) [6] [7].  
  • “Mapping transaction flows between internal and external users to determine what access is required/not required” – Although this aligns with the principles of least privilege, it is too static for today’s dynamic supply chains and evolving digital infrastructure. This approach also suggests the existence of “trusted” access routes into a network.

Attack Overview

In July 2024, Darktrace / EMAIL™ detected and contained a sophisticated phishing attack leveraging Microsoft SharePoint. This attack exploited the trusted relationship between a Darktrace customer in the public transport sector and a compromised supplier. Traditional methods, such as those detailed above, would likely have failed to defend against such an advanced threat. However, Darktrace’s behavioral analysis and zero-trust approach to email security allowed it to successfully identify and neutralize the attack, preventing any potential disruption.

Initial Intrusion Attempt

The observed phishing attack by Darktrace would suggest that the customer’s supplier was targeted by a similar campaign beforehand. This initial breach likely allowed the attacker to use the now compromised account as a vector to compromise additional accounts and networks.

On July 9, Darktrace / EMAIL identified a significant spike in inbound emails from “supplier@engineeringcompany[.]com”. The emails appeared to be legitimate notifications sent via SharePoint and contained a file named “Payment Applications Docs”.

Email correspondence in the weeks around the phishing attack.
Figure 1: Email correspondence in the weeks around the phishing attack. The sender is an established correspondent with ongoing communications prior to and after the attack, however there is a significant spike in incoming emails on the day of the attack.

This reflects a common technique in malicious social engineering attempts, where references to payment are used to draw attention and prompt a response. Darktrace observed a large number of recipients within the organization receiving the same file, suggesting that the motive was likely credential harvesting rather than financial gain. Financially motivated attacks typically require a more targeted, ‘under-the-radar’ approach to be successful.

These phishing emails were able to bypass the customer’s email gateways as they were sent from a trusted and authoritative source, SharePoint, and utilized an email address with which the customer had previously corresponded. The compromised account was likely whitelisted by traditional email security tools that rely on SPF, DKIM, and DMAC, allowing the malicious emails to evade detection.

Autonomous Response

Darktrace / EMAIL analysis of the unusual characteristics of the phishing email in relation to the supplier’s typical behaviour, despite the email originating from a legitimate SharePoint notification.
Figure 2: Darktrace / EMAIL analysis of the unusual characteristics of the phishing email in relation to the supplier’s typical behavior, despite the email originating from a legitimate SharePoint notification.

However, Darktrace / EMAIL did not use these static rules to automatically trust the email. Darktrace’s Self-Learning AI detected the following anomalies:

  • Although the sender was known, it was not normal for the supplier to share files with the customer via SharePoint.
  • The supplier initiated an unusually large number of file shares in a short period of time, indicating potential spam activity.
  • The SharePoint link had wide access permissions, which is unusual for a sensitive payment document legitimately shared between established contacts.

Darktrace understood that the email activity constituted a significant deviation in expected behavior between the sender and customer, regardless of the known sender and use of a legitimate filesharing platform like SharePoint.

As a result, Darktrace took action to hold more than 100 malicious emails connected to the phishing attack, preventing them from landing in recipient inboxes in the first instance.  By taking a behavioral approach to securing customer email environments, Darktrace’s Self-Learning AI embodies the principles of zero trust, assessing each interaction in real-time against a user’s dynamic baseline rather than relying on static and often inaccurate rules to define trust.

Conclusion

Cloud services, such as SharePoint, offer significant advantages to the transportation sector by streamlining data exchange with supply chain partners and facilitating access to information for analytics and planning. However, these benefits come with notable risks. If a cloud account is compromised, unauthorized access to sensitive information could lead to extortion and lateral movement into mission-critical systems for more damaging attacks on CNI. Even a brief disruption in cloud access can have severe economic repercussions due to the sector’s dependence on these services for resource coordination and the cascading impacts on other critical systems [9].

While supply chain resilience is often evaluated based on a supplier’s initial compliance with baseline standards, organizations must be wary of potential future threats and focus on post-implementation security. It is essential for organizations to employ strategies to protect their assets from attacks that would exploit vulnerabilities within the trusted supply chain. Given that CNI and the transportation sector are prime targets for state-sponsored actors and Advanced Persistent Threat (APT) groups, the complex and interconnected nature of their supply chains opens the door for opportunistic attackers.

Defenders face the challenge of ensuring secure access and collaboration across numerous, dynamic assets, often without full visibility. Therefore, security solutions must be as dynamic as the threats they face, avoiding reliance on static rules. Real-time assessment of devices behavior, even if deemed trusted by end-users and human security teams, is crucial for maintaining security.

Darktrace’s AI-driven threat detection aligns with the zero-trust principle of assuming the risk of a breach. By leveraging AI that learns an organization’s specific patterns of life, Darktrace provides a tailored security approach ideal for organizations with complex supply chains.

Credit to Nicole Wong, Senior Cyber Analyst Consultant and Ryan Traill, Threat Content Lead

Appendices

Darktrace Model Detections

Key model alerts:

  • Personalized Sharepoint Share + New Unknown Link
  • Personalized Sharepoint Share + Bad Display Text
  • Personalized Sharepoint Share + Distant Recipient Interaction with Domain
  • Personalized Sharepoint Share + Sender Surge
  • Personalized Sharepoint Share + Wide Access Sharepoint Link

MITRE ATT&CK Mapping

Resource Development • Compromise Accounts: Cloud Accounts • T1586.003

Initial Access • Supply Chain Compromise • T1195

References

[1] https://www.cisa.gov/topics/critical-infrastructure-security-and-resilience/critical-infrastructure-sectors

[2]  https://committees.parliament.uk/writtenevidence/126313/pdf/

[3] https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-161r1.pdf

[4] https://cloudsecurityalliance.org/press-releases/2023/11/15/cloud-security-alliance-launches-the-industry-s-first-authoritative-zero-trust-training-and-credential-the-certificate-of-competence-in-zero-trust-cczt

[5] https://www.gartner.com/en/documents/5286863#:~:text=Summary,anticipate%20staffing%20and%20cost%20increases.

[6] https://darktrace.com/threat-report-2023

[7] https://darktrace.com/resources/first-6-half-year-threat-report-2024

[8] https://dfrlab.org/2023/07/10/critical-infrastructure-and-the-cloud-policy-for-emerging-risk/#transportation

[9] https://access-national-risk-register.service.cabinetoffice.gov.uk/risk-scenario/cyber-attack-transport-sector

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
Nicole Wong
Cyber Security Analyst

More in this series

No items found.

Blog

/

/

September 24, 2026

Detecting Rogue Agent Behavior in the Enterprise

Default blog imageDefault blog image

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

Key Takeaways:

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

Introduction: Understanding the Threat of Hacking by Agents

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

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

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

Setting the Stage

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

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

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

Inducing Cheating Behavior

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

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

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

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

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

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

Details from a Daybreak Test

Continue reading
About the author
Sam Lister
Specialist Security Researcher

Blog

/

/

September 24, 2026

Agent Hijacks: Hijacking Agentic Harnesses to Attack an Organization

Default blog imageDefault blog image

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

Key takeaways:

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

Introduction: Agentic harnesses, trust, and conversation history poisoning

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

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

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

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

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

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

How we conducted the research

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

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

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

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

Agent Guardrails and Discretion

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

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

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

Agent hijack

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

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

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

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

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

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

Secure AI: Conversation history poisoning and beyond

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

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

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

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

Footnotes

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

[related-resource]

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