Blog
/
Email
/
April 10, 2023

Employee-Conscious Email Security Solutions in the Workforce

Email threats commonly affect organizations. Read Darktrace's expert insights on how to safeguard your business by educating employees about email security.
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
Dan Fein
VP, Product
Written by
Carlos Gray
Senior Product Marketing Manager, Email
Default blog image
10
Apr 2023

When considering email security, IT teams have historically had to choose between excluding employees entirely, or including them but giving them too much power and implementing unenforceable, trust-based policies that try to make up for it. 

However, just because email security should not rely on employees, this does not mean they should be excluded entirely. Employees are the ones interacting with emails daily, and their experiences and behaviors can provide valuable security insights and even influence productivity. 

AI technology supports employee engagement in this non-intrusive, nuanced way to not only maintain email security, but also enhance it. 

Finding a Balance of Employee Involvement in Security Strategies

Historically, security solutions offered ‘all or nothing’ approaches to employee engagement. On one hand, when employees are involved, they are unreliable. Employees cannot all be experts in security on top of their actual job responsibilities, and mistakes are bound to happen in fast-paced environments.  

Although there have been attempts to raise security awareness, they often have shortcomings, as training emails lack context and realism, leaving employees with poor understandings that often lead to reporting emails that are actually safe. Having users constantly triaging their inboxes and reporting safe emails wastes time that takes away from their own productivity as well as the productivity of the security team.

Other historic forms of employee involvement also put security at risk. For example, users could create blanket rules through feedback, which could lead to common problems like safe-listing every email that comes from the gmail.com domain. Other times, employees could choose for themselves to release emails without context or limitations, introducing major risks to the organization. While these types of actions include employees to participate in security, they do so at the cost of security. 

Even lower stakes employee involvement can prove ineffective. For example, excessive warnings when sending emails to external contacts can lead to banner fatigue. When employees see the same warning message or alert at the top of every message, it’s human nature that they soon become accustomed and ultimately immune to it.

On the other hand, when employees are fully excluded from security, an opportunity is missed to fine-tune security according to the actual users and to gain feedback on how well the email security solution is working. 

So, both options of historically conventional email security, to include or exclude employees, prove incapable of leveraging employees effectively. The best email security practice strikes a balance between these two extremes, allowing more nuanced interactions that maintain security without interrupting daily business operations. This can be achieved with AI that tailors the interactions specifically to each employee to add to security instead of detracting from it. 

Reducing False Reports While Improving Security Awareness Training 

Humans and AI-powered email security can simultaneously level up by working together. AI can inform employees and employees can inform AI in an employee-AI feedback loop.  

By understanding ‘normal’ behavior for every email user, AI can identify unusual, risky components of an email and take precise action based on the nature of the email to neutralize them, such as rewriting links, flattening attachments, and moving emails to junk. AI can go one step further and explain in non-technical language why it has taken a specific action, which educates users. In contrast to point-in-time simulated phishing email campaigns, this means AI can share its analysis in context and in real time at the moment a user is questioning an email. 

The employee-AI feedback loop educates employees so that they can serve as additional enrichment data. It determines the appropriate levels to inform and teach users, while not relying on them for threat detection. 

In the other direction, the AI learns from users’ activity in the inbox and gradually factors this into its decision-making. This is not a ‘one size fits all’ mechanism – one employee marking an email as safe will never result in blanket approval across the business – but over time, patterns can be observed and autonomous decision-making enhanced.  

Figure 1: The employee-AI feedback loop increases employee understanding without putting security at risk.

The employee-AI feedback loop draws out the maximum potential benefits of employee involvement in email security. Other email security solutions only consider the security team, enhancing its workflow but never considering the employees that report suspicious emails. Employees who try to do the right thing but blindly report emails never learn or improve and end up wasting their own time. By considering employees and improving security awareness training, the employee-AI feedback loop can level up users. They learn from the AI explanations how to identify malicious components, and so then report fewer emails but with greater accuracy. 

While AI programs have classically acted like black boxes, Darktrace trains its AI on the best data, the organization’s actual employees, and invites both the security team and employees to see the reasoning behind its conclusions. Over time, employees will trust themselves more as they better learn how to discern unsafe emails. 

Leveraging AI to Generate Productivity Gains

Uniquely, AI-powered email security can have effects outside of security-related areas. It can save time by managing non-productive email. As the AI constantly learns employee behavior in the inbox, it becomes extremely effective at detecting spam and graymail – emails that aren't necessarily malicious, but clutter inboxes and hamper productivity. It does this on a per-user basis, specific to how each employee treats spam, graymail, and newsletters. The AI learns to detect this clutter and eventually learns which to pull from the inbox, saving time for the employees. This highlights how security solutions can go even further than merely protecting the email environment with a light touch, to the point where AI can promote productivity gains by automating tasks like inbox sorting.

Preventing Email Mishaps: How to Deal with Human Error

Improved user understanding and decision making cannot stop natural human error. Employees are bound to make mistakes and can easily send emails to the wrong people, especially when Outlook auto-fills the wrong recipient. This can have effects ranging anywhere from embarrassing to critical, with major implications on compliance, customer trust, confidential intellectual property, and data loss. 

However, AI can help reduce instances of accidentally sending emails to the wrong people. When a user goes to send an email in Outlook, the AI will analyze the recipients. It considers the contextual relationship between the sender and recipients, the relationships the recipients have with each other, how similar each recipient’s name and history is to other known contacts, and the names of attached files.  

If the AI determines that the email is outside of a user’s typical behavior, it may alert the user. Security teams can customize what the AI does next: it can block the email, block the email but allow the user to override it, or do nothing but invite the user to think twice. Since the AI analyzes each email, these alerts are more effective than consistent, blanket alerts warning about external recipients, which often go ignored. With this targeted approach, the AI prevents data leakage and reduces cyber risk. 

Since the AI is always on and continuously learning, it can adapt autonomously to employee changes. If the role of an employee evolves, the AI will learn the new normal, including common behaviors, recipients, attached file names, and more. This allows the AI to continue effectively flagging potential instances of human error, without needing manual rule changes or disrupting the employee’s workflow. 

Email Security Informed by Employee Experience

As the practical users of email, employees should be considered when designing email security. This employee-conscious lens to security can strengthen defenses, improve productivity, and prevent data loss.  

In these ways, email security can benefit both employees and security teams. Employees can become another layer of defense with improved security awareness training that cuts down on false reports of safe emails. This insight into employee email behavior can also enhance employee productivity by learning and sorting graymail. Finally, viewing security in relation to employees can help security teams deploy tools that reduce data loss by flagging misdirected emails. With these capabilities, Darktrace/Email™ enables security teams to optimize the balance of employee involvement in email security.

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
Dan Fein
VP, Product
Written by
Carlos Gray
Senior Product Marketing Manager, Email

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