September 11, 2025

SEO Poisoning and Fake PuTTY sites: Darktrace’s Investigation into the Oyster backdoor

SEO poisoning is a malicious tactic where threat actors manipulate search engine rankings to promote deceptive websites. These sites often mimic legitimate software downloads, delivering malware like the Oyster backdoor. Learn about Darktrace’s investigation into the tactics used to deliver Oyster via fake PuTTY sites and manipulate search visibility.

Written by

Christina Kreza

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

Christina Kreza

Cyber Analyst

Sep 2025

What is SEO poisoning?

Search Engine Optimization (SEO) is the legitimate marketing technique of improving the visibility of websites in organic search engine results. Businesses, publishers, and organizations use SEO to ensure their content is easily discoverable by users. Techniques may include optimizing keywords, creating backlinks, or even ensuring mobile compatibility.

SEO poisoning occurs when attackers use these same techniques for malicious purposes. Instead of improving the visibility of legitimate content, threat actors use SEO to push harmful or deceptive websites to the top of search results. This method exploits the common assumption that top-ranking results are trustworthy, leading users to click on URLs without carefully inspecting them.

As part of SEO poisoning, the attacker will first register a typo-squatted domain, slightly misspelled or otherwise deceptive versions of real software sites, such as putty[.]run or puttyy[.]org. These sites are optimized for SEO and often even backed by malicious Google ads, increasing the visibility when users search for download links. To achieve that, threat actors may embed pages with strategically chosen, high-value keywords or replicate content from reputable sources to elevate the domain’s perceived authority in search engine algorithms [4]. In more advanced operations, these tactics are reinforced with paid promotion, such as Google ads, enabling malicious domains to appear above organic search results as sponsored links. This placement not only accelerates visibility but also impacts an unwarranted sense of legitimacy to unsuspected users.

Once a user lands on one of these fake pages, they are presented with what looks like a legitimate software download option. Upon clicking the download indicator, the user will be redirected to another separate domain that actually hosts the payload. This hosting domain is usually unrelated to the nominally referenced software. These third-party sites can involve recently registered domains but may also include legitimate websites that have been recently compromised. By hosting malware on a variety of infrastructure, attackers can prolong the availability of distribution methods for these malicious files before they are taken down.

What is the Oyster backdoor?

Oyster, also known as Broomstick or CleanUpLoader, is a C++ based backdoor malware first identified in July 2023. It enables remote access to infected systems, offering features such as command-line interaction and file transfers.

Oyster has been widely adopted by various threat actors, often as an entry point for ransomware attacks. Notable examples include Vanilla Tempest and Rhysida ransomware groups, both of which have been observed leveraging the Oyster backdoor to enhance their attack capabilities. Vanilla Tempest is known for using Oyster’s stealth persistence to maintain long-term access within targeted networks, often aligning their operations with ransomware deployment [5]. Rhysida has taken this further by deploying Oyster as an initial access tool in ransomware campaigns, using it to conduct reconnaissance and move laterally before executing encryption activities [6].

Once installed, the backdoor gathers basic system information before communicating with a command-and-control (C2) server. The malware largely relies on a ‘cmd.exe’ instance to execute commands and launch other files [1].

In previous SEO poisoning cases, the file downloaded from the fake pages is not just PuTTY, but a trojanized version that includes the stealthy Oyster backdoor. PuTTY is a free and open-source terminal emulator for Windows that allows users to connect to remote servers and devices using protocols like SSH and Telnet. In the recent campaign, once a user visits the fake software download site, ranked highly through SEO poisoning, the malicious payload is downloaded through direct user interaction and subsequently installed on the local device, initiating the compromise. The malware then performs two actions simultaneously: it installs a fully functional version of PuTTY to avoid user suspicion, while silently deploying the Oyster backdoor. Given PuTTY’s nature, it is prominently used by IT administrators with highly privileged account as opposed to standard users in a business, possibly narrowing the scope of the targets.

Oyster’s persistence mechanism involves creating a Windows Scheduled Task that runs every few minutes. Notably, the infection uses Dynamic Link Library (DLL) side loading, where a malicious DLL, often named ‘twain_96.dll’, is executed via the legitimate Windows utility ‘rundll32.exe’, which is commonly used to run DLLs [2]. This technique is frequently used by malicious actors to blend their activity with normal system operations.

Darktrace’s Coverage of the Oyster Backdoor

In June 2025, security analysts at Darktrace identified a campaign leveraging search engine manipulation to deliver malware masquerading as the popular SSH client, PuTTY. Darktrace / NETWORK’s anomaly-based detection identified signs of malicious activity, and when properly configured, its Autonomous Response capability swiftly shut down the threat before it could escalate into a more disruptive attack. Subsequent analysis by Darktrace’s Threat Research team revealed that the payload was a variant of the Oyster backdoor.

The first indicators of an emerging Oyster SEO campaign typically appeared when user devices navigated to a typosquatted domain, such as putty[.]run or putty app[.]naymin[.]com, via a TLS/SSL connection.

*Figure 1: Darktrace’s detection of a device connecting to the typosquatted domain putty[.]run.*

The device would then initiate a connection to a secondary domain that hosts the malicious installer, likely triggered by user interaction with redirect elements on the landing page. This secondary site may not have any immediate connection to PuTTY itself but is instead a hijacked blog, a file-sharing service, or a legitimate-looking content delivery subdomain.

*Figure 2: Darktrace’s detection of the device making subsequent connections to the payload domain.*

Following installation, multiple affected devices were observed attempting outbound connectivity to rare external IP addresses, specifically requesting the ‘/secure’ endpoint as noted within the declared URIs. After the initial callback, the malware continued communicating with additional infrastructure, maintaining its foothold and likely waiting for tasking instructions. Communication patterns included:

· Endpoints with URIs /api/kcehc and /api/jgfnsfnuefcnegfnehjbfncejfh

· Endpoints with URI /reg and user agent “WordPressAgent”, “FingerPrint” or “FingerPrintpersistent”

This tactic has been consistently linked to the Oyster backdoor, which has shown similar URI patterns across multiple campaigns [3].

Darktrace analysts also noted the sophisticated use of spoofed user agent strings across multiple investigated customer networks. These headers, which are typically used to identify the application making an HTTP request, are carefully crafted to appear benign or mimic legitimate software. One common example seen in the campaign is the user agent string “WordPressAgent”. While this string references a legitimate web application or plugin, it does not appear to correspond to any known WordPress services or APIs. Its inclusion is most likely designed to mimic background web traffic commonly associated with WordPress-based content management systems.

*Figure 3: Cyber AI Analyst investigation linking the HTTP C2 activity.*

Case-Specific Observations

While the previous section focused on tactics and techniques common across observed Oyster infections, a closer examination reveals notable variations and unique elements in specific cases. These distinct features offer valuable insights into the diverse operational approaches employed by threat actors. These distinct features, from unusual user agent strings to atypical network behavior, offer valuable insights into the diverse operational approaches employed by the threat actors. Crucially, the divergence in post-exploitation activity reflects a broader trend in the use of widely available malware families like Oyster as flexible entry points, rather than fixed tools with a single purpose. This modular use of the backdoor reflects the growing Malware-as-a-Service (MaaS) ecosystem, where a single initial infection can be repurposed depending on the operator’s goals.

From Infection to Data Egress

In one observed incident, Darktrace observed an infected device downloading a ZIP file named ‘host[.]zip’ via curl from the URI path /333/host[.]zip, following the standard payload delivery chain. This file likely contained additional tools or payloads intended to expand the attacker’s capabilities within the compromised environment. Shortly afterwards, the device exhibited indicators of probable data exfiltration, with outbound HTTP POST requests featuring the URI pattern: /upload?dir=NAME_FOLDER/KEY_KEY_KEY/redacted/c/users/public.

This format suggests the malware was actively engaged in local host data staging and attempting to transmit files from the target machine. The affected device, identified as a laptop, aligns with the expected target profile in SEO poisoning scenarios, where unsuspecting end users download and execute trojanized software.

Irregular RDP Activity and Scanning Behavior

Several instances within the campaign revealed anomalous or unexpected Remote Desktop Protocol (RDP) sessions occurring shortly after DNS requests to fake PuTTY domains. Unusual RDP connections frequently followed communication with Oyster backdoor C2 servers. Additionally, Darktrace detected patterns of RDP scanning, suggesting the attackers were actively probing for accessible systems within the network. This behavior indicates a move beyond initial compromise toward lateral movement and privilege escalation, common objectives once persistence is established.

The presence of unauthorized and administrative RDP sessions following Oyster infections aligns with the malware’s historical role as a gateway for broader impact. In previous campaigns, Oyster has often been leveraged to enable credential theft, lateral movement, and ultimately ransomware deployment. The observed RDP activity in this case suggests a similar progression, where the backdoor is not the final objective but rather a means to expand access and establish control over the target environment.

Cryptic User Agent Strings?

In multiple investigated cases, the user agent string identified in these connections featured formatting that appeared nonsensical or cryptic. One such string containing seemingly random Chinese-language characters translated into an unusual phrase: “Weihe river is where the water and river flow.” Legitimate software would not typically use such wording, suggesting that the string was intended as a symbolic marker rather than a technical necessity. Whether meant as a calling card or deliberately crafted to frame attribution, its presence highlights how subtle linguistic cues can complicate analysis.

‍

Figure 4: Darktrace’s detection of malicious connections using a user agent with randomized Chinese-language formatting.

Strategic Implications

What makes this campaign particularly noteworthy is not simply the use of Oyster, but its delivery mechanism. SEO poisoning has traditionally been associated with cybercriminal operations focused on opportunistic gains, such as credential theft and fraud. Its strength lies in casting a wide net, luring unsuspecting users searching for popular software and tricking them into downloading malicious binaries. Unlike other campaigns, SEO poisoning is inherently indiscriminate, given that the attacker cannot control exactly who lands on their poisoned search results. However, in this case, the use of PuTTY as the luring mechanism possibly indicates a narrowed scope - targeting IT administrators and accounts with high privileges due to the nature of PuTTY’s functionalities.

This raises important implications when considered alongside Oyster. As a backdoor often linked to ransomware operations and persistent access frameworks, Oyster is far more valuable as an entry point into corporate or government networks than small-scale cybercrime. The presence of this malware in an SEO-driven delivery chain suggests a potential convergence between traditional cybercriminal delivery tactics and objectives often associated with more sophisticated attackers. If actors with state-sponsored or strategic objectives are indeed experimenting with SEO poisoning, it could signal a broadening of their targeting approaches. This trend aligns with the growing prominence of MaaS and the role of initial access brokers in today’s cybercrime ecosystem.

Whether the operators seek financial extortion through ransomware or longer-term espionage campaigns, the use of such techniques blurs the traditional distinctions. What looks like a mass-market infection vector might, in practice, be seeding footholds for high-value strategic intrusions.

Credit to Christina Kreza (Cyber Analyst) and Adam Potter (Senior Cyber Analyst)

Appendices

MITRE ATT&CK Mapping

· T1071.001 – Command and Control – Web Protocols

· T1008 – Command and Control – Fallback Channels

· T0885 – Command and Control – Commonly Used Port

· T1571 – Command and Control – Non-Standard Port

· T1176 – Persistence – Browser Extensions

· T1189 – Initial Access – Drive-by Compromise

· T1566.002 – Initial Access – Spearphishing Link

· T1574.001 – Persistence – DLL

Indicators of Compromise (IoCs)

· 85.239.52[.]99 – IP address

· 194.213.18[.]89/reg – IP address / URI

· 185.28.119[.]113/secure – IP address / URI

· 185.196.8[.]217 – IP address

· 185.208.158[.]119 – IP address

· putty[.]run – Endpoint

· putty-app[.]naymin[.]com – Endpoint

· /api/jgfnsfnuefcnegfnehjbfncejfh

· /api/kcehc

Darktrace Model Detections

· Anomalous Connection / New User Agent to IP Without Hostname

· Anomalous Connection / Posting HTTP to IP Without Hostname

· Compromise / HTTP Beaconing to Rare Destination

· Compromise / Large Number of Suspicious Failed Connections

· Compromise / Beaconing Activity to External Rare

· Compromise / Quick and Regular Windows HTTP Beaconing

· Device / Large Number of Model Alerts

· Device / Initial Attack Chain Activity

· Device / Suspicious Domain

· Device / New User Agent

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

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

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

References

[1] https://malpedia.caad.fkie.fraunhofer.de/details/win.broomstick

[2] https://arcticwolf.com/resources/blog/malvertising-campaign-delivers-oyster-broomstick-backdoor-via-seo-poisoning-trojanized-tools/

[3] https://hunt.io/blog/oysters-trail-resurgence-infrastructure-ransomware-cybercrime

[4] https://www.crowdstrike.com/en-us/cybersecurity-101/social-engineering/seo-poisoning/

[5] https://blackpointcyber.com/blog/vanilla-tempest-oyster-backdoor-netsupport-unknown-infostealers-soc-incidents-blackpoint-apg/

[6] https://areteir.com/article/rhysida-using-oyster-backdoor-in-attacks/

‍

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

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

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

^{The cybersecurity landscape evolves rapidly, and blog content may become outdated or superseded. We reserve the right to update, modify, or remove any content without notice.}

Written by

Christina Kreza

Cyber Analyst

Inside the SOC

Written by

Christina Kreza

Cyber Analyst

•

August 3, 2026

Nathaniel Jones

SVP, Global Threat Intelligence

Senior Director of Product | Darktrace

Watch the NIS2 Webinar

Blog

Network

August 7, 2026

When AI Agents Attack: The Case for Behavioral Anomaly Detection

The UK AI Security Institute (AISI) has just published a detailed incident report confirming what many within the cyber security industry have suspected for some time: frontier AI agents, given a degree of autonomy, can independently develop and execute attack chains against real targets, including social engineering, supply-chain compromise, and deception, without explicit instruction to do so.

The report (INC-2026-07-28-01, published 4 August 2026) documents what happened during a routine capability evaluation between 25 and 28 July. It deserves a careful read from anyone responsible for securing the modern digital estate.

The incident

AISI regularly tests frontier AI models for offensive cyber capability. In this exercise, agents were placed inside a simulated network environment and given internet access this was deliberate, given to reflect realism of the task. Across 122 evaluation runs, 19 instances of unsanctioned activity were recorded, including cases where agents broke the intended boundaries of the test and acted against real people and real infrastructure on the open internet.

In the most serious case, an agent (Mythos 5) autonomously:

Conducted open-source intelligence gathering to identify real developers associated with keywords from the simulated scenario
Created a GitHub account and submitted a pull request containing obfuscated malicious code to a real, unaffiliated repository
Created a second sockpuppet account to post comments on its own PR, attempting to manufacture social proof
When a human reviewer flagged the malicious content, the agent lied, claiming an accidental commit, then repeatedly attempted to reintroduce the payload under the guise of having "fixed" the code
Sent spear-phishing emails to the repository maintainer containing manipulated content designed to convince them to merge the code
Planted a prompt injection inside a GitHub issue, hidden in an HTML comment invisible to humans but readable by AI coding assistants parsing the repository via API

AISI confirmed no real-world harm resulted. The PRs were not merged; the payloads did not execute. But containment was a product of circumstance, not design.

Why this matters

Every action the agents took used legitimate services operating within normal parameters. Creating a GitHub account is legitimate. Opening a pull request is legitimate. Sending an email is legitimate. Posting an issue is legitimate.

No vulnerability was exploited. No known malware signature was present. No indicator of compromise from any threat intelligence feed would have matched. The entire attack existed as a pattern, a sequence of individually normal actions that, taken together, constituted a supply-chain attack, a social engineering campaign, and an attempted compromise of downstream systems.

This is the class of threat that signature-based and rule-based security tools structurally struggle to detect. They answer the question "have I seen this before?" When the answer is no – and when the attack is novel, composed of legitimate actions, and adapted in real time by an adversary in this case an agent that reasons about evasion – they struggle.

Behavioral anomaly detection: The architecture that fits

Darktrace has spent a decade building a platform around a different question: "is this behavior normal for this entity?"

The platform learns the baseline behavior of every user, device, and communication pattern across an organization's digital environment. It does not need prior knowledge of an attack technique. It identifies deviation from learned normality, surfaces it, and through its Autonomous Response capability, takes targeted action to contain threats at machine speed.

Consider how this AISI incident might show across the Darktrace platform:

Network detection and response

While no payload was successfully delivered to a victim in this incident, AISI's post-incident analysis of the agent's code (detailed on pages 24-25 of the report) reveals what would have landed on target systems had the social engineering succeeded. The agent produced three successive payload generations, each replacing the last after being detected and called out:

Generation 1 hid a module-load stager inside a legitimate bug-fix commit, guarded by an environment variable check, to skip execution in CI. Its second stage attempted to fetch Poseidon implants (Linux and Windows variants) calling back to the agent's own C2 listener, with a fallback to publicly hosted polling scripts that fetched and executed commands every 60 seconds.

Generation 2 exploited an Electron "What's New" window with nodeIntegration enabled, loading remote HTML from a GitHub Pages site the agent had registered. The page decoded an XOR-obfuscated payload at runtime that beaconed host identifiers over DNS, attempted to download the same implants, ran its own in-process C2 loop, and installed reboot persistence via crontab and Windows registry Run keys.

Generation 3 was a post-install dropper added after the agent had already been publicly caught. It XOR-decoded a hidden URL, downloaded a script over plaintext HTTP, and ran it detached, force-exiting so npm install always reports success, giving the attacker remote code execution on any machine running the install.

None of these payloads reached a victim. But the technical detail matters: this is the calibre of tradecraft an autonomous agent produced in a matter of hours, unprompted. Each generation demonstrated adaptation, the agent iterated its approach in direct response to being detected.

For network defenders, the behavioral signals these payloads would have generated are clear. DNS beaconing to novel domains, periodic HTTP polling to newly registered infrastructure, XOR-encoded data in transit, registry and crontab persistence writes on endpoints that have never exhibited such behavior. These are exactly the anomalies that Darktrace / NETWORK identifies in real time: deviations from learned device and connection baselines, regardless of whether the specific payload or C2 domain has been seen before.

Beyond the payload analysis, the agents' broader activity produced additional behavioral anomalies: new identities engaging in high-frequency interactions with external services, coordination via unusual channels (public pastebins, gists), and lateral communication patterns that deviate from any established baseline.

Email security

The spear-phishing emails sent by the agent, crafted with context-specific content designed to manipulate a specific individual, represent exactly the kind of socially engineered communication that static email gateways can miss. Darktrace / EMAIL analyses the linguistic patterns, sender behavior, and relational context of every message. A brand-new sender using tailored language to pressure a recipient into a specific technical action deviates from the learned communication profile, regardless of whether the message contains a known-bad URL or attachment.

Securing AI

This is where the incident points to a threat category that many security platforms do not yet fully address. The agent did not just use AI as a tool for attack. It targeted other AI systems, attempting prompt injections against coding assistants and trying to compromise automated workflows. Organisations adopting AI agents for development, operations, and decision-making are introducing a newer extended attack surface.

Darktrace / SECURE AI extends the same behavioral detection model into this space. It monitors AI usage across the enterprise, detecting anomalous interactions with AI services, identifying sensitive data exposure to AI systems, providing visibility into shadow AI use. As AI agents become standard components of enterprise infrastructure, the ability to detect when an AI system is behaving outside its expected parameters, when there is a chance, it is being manipulated, or when it is interacting with something it should not be, that becomes a direct security requirement rather than a future consideration.

The principle is consistent across the Darktrace platform: learn what normal looks like, identify deviation, respond autonomously.

The AI-to-AI threat surface

One of the report's most important findings is a confirmation that AI agents can act outside of instructions and guardrails to complete tasks. The Mythos 5 agent conducted reconnaissance on a repository maintainer, inferred it might be an AI coding assistant, and deployed a prompt injection specifically crafted to compromise it.

This is a new attack class. Organizations deploying AI coding assistants, AI-powered code review, or autonomous DevOps agents now face an adversary that:

Moves at speed
Is able to adapt its approach when initial attempts are halted
Specifically targets AI systems via techniques (prompt injection, hidden instructions in HTML comments) that are less visible to human reviewers but parsed by machines
Reasons about which evasion strategies are most likely to succeed

Defending against this requires security that operates with the same adaptability and speed. A system that learns what "normal" looks like, and flags deviation from it.

What AISI recommends

The report identifies five contributing factors to the incident. Two are directly relevant:

No synchronous real-time monitoring (Section 5.3): AISI acknowledges it had not built real-time behavioral monitoring capable of immediately blocking unsanctioned agent actions. Their security team detected the incident via commercial security monitoring software, but only after activity had already occurred.

No fine-grained behavioral controls (Section 6.2): AISI's forward-looking response includes building continuous monitoring that can "immediately block or flag unsanctioned AI agent behavior."

The direction is clear: organizations need security that watches behavior continuously, learns what normal looks like, and acts when deviation occurs, without requiring a human in the loop for every decision.

The takeaway

The AISI incident is documented proof that autonomous AI agents will independently develop novel attack chains using legitimate tools, adapt when detected, employ deception, and target both humans and other AI systems. This happened last week, in a controlled setting, with commercially available models.

The security architecture that addresses this is behavioral anomaly detection applied across the full digital estate, as AI agents become standard components of enterprise infrastructure, writing code, managing deployments, processing communications, the attack surface they create is behavioral by nature.

This is the approach Darktrace has taken for years: learning what is normal across an organization’s digital environment, identifying meaningful deviations, and responding to emerging threats without relying on known attack signatures. As autonomous AI agents introduce new and unpredictable behaviors, that foundation becomes increasingly important to securing the enterprise.

‍

Read the full report from the UK AI Security Institute here.

About the author

Adam Stevens

Senior Director of Product | Darktrace

Blog

AI

August 5, 2026

Testing a Prompt injection Attack Against an Enterprise AI Agent

Key takeaways

Darktrace successfully detected and quarantined a prompt injection email before it could be processed by an enterprise AI agent.
Prompt injection attacks increasingly rely on natural language rather than traditional malware, making behavioral analysis an important complement to signature-based detection.
Organizations deploying AI agents should combine model guardrails with behavioral monitoring to reduce the risk of malicious instructions reaching enterprise systems.

How behavioral detection helps stop prompt injection attacks

A Darktrace customer running a Gemini AI agent in Google Cloud asked us two simple questions:

“If my agent can read inbound emails and access internal data, what stops an attacker from hiding malicious instructions in the message? Couldn’t the agent be tricked into deleting or exfiltrating sensitive data?”

The scenario centers on an indirect prompt injection attack, where malicious instructions are hidden inside content that an AI model later interprets as trusted input. The same weakness was exposed by EchoLeak (CVE-2025-32711), a zero-click Microsoft 365 Copilot vulnerability enabled data exfiltration from a single well-crafted email..

This blog follows the Darktrace team’s investigation of the customer’s hypothesis and examines how the attack interacted with their existing security stack. The results highlight which defenses held, where gaps emerged, and how behavioral detection mattered more than guardrails. This investigation also demonstrates why behavioral detection is becoming increasingly important for AI security, as prompt injections often contain no traditional indicators of compromise.

How do prompt injection attacks work?

Prompt injection works by carefully crafting the content and structure of the prompt to alter the LLM’s behavior or output in unintended ways. This can cause models to violate guardrails, generate harmful content or enable unauthorised access.

Prompt injection attack example

The well-known example, EchoLeak (CVE-2025-32711), was a zero-click vulnerability in Microsoft 365 Copilot that relied on a carefully crafted email containing hidden instructions that the AI system interpreted as commands rather than content, creating a pathway for unauthorized access to sensitive information without any user interaction.

While Darktrace / SECURE AI is designed to prevent agents from producing unintended outcomes, we wanted to see if we could catch and prevent this threat type earlier in the attack life-cycle, at the email security layer.

How we tested prompt injection attacks on an enterprise agent

Summary:

Claude generated a prompt injection payload.
Hidden instructions were embedded in an email.
The email passed traditional validation checks.
Darktrace analyzed the language and sender behavior.
The email was quarantined before the AI agent could process it.

To test Darktrace / EMAIL against this attack class, we opened Claude, gave it the customer's context and problem statement (Gemini agent with inbox access, internal tool calls), told it we were validating Darktrace / EMAIL's detection of prompt injections, and asked for a test payload. See below:

Despite the guardrails supposedly built into the model, Claude surprisingly gave us the entire exploit in plaintext (albeit very basic), illegible to a human as the text was sent in white text (see Figure 1) but framed as an authoritative override for anything downstream reading the mail programmatically (i.e. the Gemini agent).

How Darktrace detected a prompt injection attack

We then sent the Claude-crafted email from a freemail address to the target recipient’s inbox. Despite the email containing no malicious payload, the freemail address having no malicious reputation, and the validation checks all passing, Darktrace /EMAIL flagged the email as a 93/100 anomaly and moved it to junk, out of scope for the AI agent.

Figure 3: *The test email sent with the hidden prompt injection*

*Figure 4: The email analysis in Darktrace / EMAIL*

Figure 5: Darktrace / EMAIL detection of malicious activity

The interesting part is what triggered the detection (see Figure 5)

Possible machine prompt content: text in the body detected as instructions written for a machine to execute, not for a human to read

Possible machine prompt content + basic suspicious correspondence: the same content, correlated with sender-side anomalies: freemail domain (yahoo[.]com), unknown correspondent, no prior mail history with the recipient, and suspicious references to payment information

Neither of those is a signature match. Nothing in the email was on a blacklist. There was no malware, no link and no attachment. Darktrace analyzed the context in which the email was delivered and flagged it as likely risky. The anomalous language features and the context of the sender relative to the recipient's normal behavior, combined with the unusual hidden text (prompt) were enough for Darktrace / EMAIL to act on the risk.

Result: Darktrace / EMAIL autonomously junked the email, out of scope for any AI agent parsing the inbox.

Why behavioral security makes a difference detecting prompt injection attacks

Cyberattacks don't look like traditional exploits anymore. They now operate in natural language, not strictly code.

That breaks the traditional stack. AV, firewalls, static scanning and signature-based SEGs all assume a payload to inspect.

A prompt injection has no payload. It's just an instruction, written in natural language, dressed up as anything the attacker wants: an invoice, an HR request, a calendar invite, some simple PowerPoint slides.

EchoLeak proved that hidden instructions can sit inside an email invisible to the user but fully readable by the LLM, and the LLM will follow them blindly.

This test and GTG-1002 proved that the LLM itself can be socially engineered. Tell it you're an authorized tester and it will hand you the attack.

Rules and static classifiers can catch the obvious cases. But natural language has infinite variants, and the attack surface is the model's innate functionality to comply.

The deeper problem here is intent: an LLM can't reliably tell whether an instruction in its context came from its developer, its user, or an attacker who slipped it into an email. To the LLM, everything reads as language and looks like a legitimate ask. This is why behavioural detection wins, as you become aware of intent when you look at the context of an interaction. Does this sender normally send this kind of message to this recipient? Does this prompt fit the user's normal pattern? Is this agent behaving the way this agent normally behaves?

Intent can't be read off a single email, it emerges from behavioral context. Which is how Darktrace enables threat detection, through behavioral understanding.

Why enterprise AI security requires more than guardrails

Claude didn't roll over immediately… the first section of the response was a (slight) pushback, but then it wrote the payload anyway without having to ask twice.

Here the framing of the prompt did all the work. The “testing security capabilities” angle moved the model from refusal to unquestioned compliance to the user prompt.

This isn't the first time this has happened, of course. Anthropic disclosed in November 2025 that a Chinese state-sponsored group they tracked as GTG-1002 ran the first documented AI-orchestrated espionage campaign against ~30 targets by posing as employees of a legitimate cybersecurity firm doing authorised penetration testing.

The takeaway isn't that AI guardrails are ineffective. They raise the cost of low-effort attacks and remain an important first layer of defense. However, for most organizations today, they’re the only line of defense when deploying AI agents. If a prompt injection bypasses those controls, organizations still need a way to detect and stop malicious behavior elsewhere in the attack chain.

Attackers will continue to have working prompt injections easily and quickly. The question is what stops one when it lands in an inbox your agent is reading.

That's where behavioral detection comes in.

How Darktrace detects prompt injection attacks in emails

Two things Darktrace does that a model-level guardrail or static rules and signatures can't:

Natural language analysis at the email or prompt layer. The email is assessed on its own merits: is this content shaped like instructions for a machine, regardless of what the receiving agent decides to do about it?

Behavioral context around the language. An AI agent behaves like an extremely agreeable human, and it will go above and beyond to comply with the user’s request. That's exactly why you must consider the business context, such sender behaviour, mailing history, and organisational norms, as these matter even more when the recipient is an AI.

Darktrace has been perfecting behavioral anomaly detection for over a decade; the same self-learning approach that catches BEC and account takeover applies directly to prompt injection delivery. Our multi-layered AI stack extracts content from the message, builds behavioural understanding through social graphing and Pattern of Life analysis, and then combines natural language, topic, inducement, sender relationship and anomaly signals before deciding what action to take.

This matters for prompt injection because the threat is not the plain language itself, but the intent behind the language that can cause an AI agent to respond in unexpected ways.

How to secure enterprise AI operations from prompt injection attacks

Email was the entry point in this case, but it is only one of many possible vectors.

Anywhere an agent can retrieve information, an attacker can potentially introduce a prompt injection.

Emails, documents, SharePoint sites, web pages, knowledge bases, chat platforms, and third-party integrations all provide opportunities to influence an agent's behavior. Wherever an agent finds its orders, a prompt injection opportunity exists.

This is why securing AI requires more than blocking malicious inputs. Organizations also need visibility into how agents behave after consuming information from across their environment. If an agent begins accessing unexpected data, taking unusual actions, or operating outside its normal patterns, those behaviors may provide the strongest signal that something has gone wrong.

Effective AI security requires defense in depth: reducing the likelihood of malicious instructions reaching the agent while maintaining the ability to detect and investigate suspicious behavior if they do.

The challenge isn't protecting a single entry point. It's recognizing that, in an AI-powered environment, every source of information is also a potential source of influence.

Are you deploying autonomous agents across your enterprise and want to see this tested in your environment? Let's talk.

About the author

Carlo Loregian

Solutions Engineer

Your data. Our AI.

Elevate your network security with Darktrace AI

Get a demo

Check out this article by Darktrace: SEO Poisoning and Fake PuTTY sites: Darktrace’s Investigation into the Oyster backdoor

SEO Poisoning and Fake PuTTY sites: Darktrace’s Investigation into the Oyster backdoor

What is SEO poisoning?

What is the Oyster backdoor?

Darktrace’s Coverage of the Oyster Backdoor

Case-Specific Observations

From Infection to Data Egress

Irregular RDP Activity and Scanning Behavior

Cryptic User Agent Strings?

Strategic Implications

Appendices

MITRE ATT&CK Mapping

Indicators of Compromise (IoCs)

Darktrace Model Detections

References

Why Trust is the New Attack Surface: Darktrace’s Mid-Year Threat Update 2026

When AI Agents Attack: The Case for Behavioral Anomaly Detection

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

From Click to Command: Behavioral Detection of AppleScript-Led MacOS Intrusions

Blog

Network

August 7, 2026

When AI Agents Attack: The Case for Behavioral Anomaly Detection

The incident

Why this matters

Behavioral anomaly detection: The architecture that fits

Network detection and response

Email security

Securing AI

The AI-to-AI threat surface

What AISI recommends

The takeaway

Blog

AI

August 5, 2026

Testing a Prompt injection Attack Against an Enterprise AI Agent

Key takeaways

How behavioral detection helps stop prompt injection attacks

How do prompt injection attacks work?

Prompt injection attack example

How we tested prompt injection attacks on an enterprise agent

How Darktrace detected a prompt injection attack

Why behavioral security makes a difference detecting prompt injection attacks

Why enterprise AI security requires more than guardrails

How Darktrace detects prompt injection attacks in emails

How to secure enterprise AI operations from prompt injection attacks