Blog
/
Network
/
May 19, 2026

When Open Source Is Weaponized: Analysis of a Trojanized 7 Zip Installer

Attackers abused a trojanized 7‑Zip installer distributed via a typo‑squatted domain to establish proxyware infections worldwide. This blog analyzes how social engineering, trusted open‑source software, and stealthy command‑and‑control activity enabled widespread compromise, highlighting the importance of software validation, user awareness, and proactive network‑level detection.
Written by
Justin Torres
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
Justin Torres
Cyber Analyst
7 zip installerDefault blog image
19
May 2026

Background of the malicious 7-Zip installer, and assessing its Impact

Early in 2026, external researchers disclosed a malicious distribution campaign leveraging a trojanized installer masquerading itself as a legitimate 7‑Zip utility. Evidence suggests the campaign was active as of January 2026, during which victims were served a fake installer from 7zip[.]com, a highly convincing typo-squatted domain impersonating the official 7‑Zip distribution site (7-zip[.]org).

Initial access is typically achieved through social engineering and search‑engine abuse, including YouTube tutorial content that explicitly referenced the impersonated domain as the download source. Notably, several reports observed the installer delivered a modified but functional build of 7‑Zip (7zfm.exe) to reduce suspicion and preserve expected user behavior.

However, the installer also dropped additional payloads, such as Uphero.exe, hero.exe, and hero.dll, which are not part of the legitimate 7‑Zip software package. Once installed and executed, these payloads allow the attacker to establish persistence and configure the infected host as a proxy node under their control. This facilitates malicious activities such as traffic relaying, anonymizing infrastructure, and the delivery of secondary payloads [1] [2].

Overall, this attack illustrates a proxyware-style attack that abuses implicit trust in widely deployed third‑party tools while exploiting unconventional delivery vectors such as instructional media. By closely imitating legitimate software behavior and branding, the threat actors significantly reduced user suspicion and increased the likelihood of widespread, undetected compromise.

Threat overview

Darktrace observed multiple customers affected by the malicious 7‑Zip installer between January 12 and January 22, impacting organizations across the Americas (AMS), Asia‑Pacific & Japan (APJ), and Europe, the Middle East, and Africa (EMEA) regions. The activity targeted customers across various sectors, including Human health and social work activities, Manufacturing, Education, and Information and communication.

The following use case highlights a device on one customer network making external connections associated with malicious 7-Zip update activity observed between  January 7 and January 18, 2026.  This behavior included connectivity to the malicious domain 7zip[.]com, followed by command-and control (C2) activity involving "smshero"-themed domains, as well as outbound proxy connections over ports 1000 and 1002.  

Initial Connectivity to 'update[.]7zip[.]com':

Initial Beaconing to Young Endpoint alert behavior, involving the known tunnel/proxy endpoint ‘79.127.221[.]47’.
Figure 1: Initial Beaconing to Young Endpoint alert behavior, involving the known tunnel/proxy endpoint ‘79.127.221[.]47’.

Starting on January 7, Darktrace / NETWORK detected the device making repeated beaconing connections to the endpoint 79.127.221[.]47 over the destination port 1000. The use of this port aligns with open-source intelligence (OSINT) reporting that hero[.]exe establishes outbound proxy connections via non-standard ports such as 1000 and 1002 [1].

Darktrace observed TLS beaconing alerts to the known trojanized installer, update[.]7zip[.]com · 98.96.229[.]19, over port 443 on January 7th.
Figure 2: Darktrace observed TLS beaconing alerts to the known trojanized installer, update[.]7zip[.]com · 98.96.229[.]19, over port 443 on January 7th.

Later the same day, the device initiated TLS beaconing to the endpoint update.7zip[.]com. This is more than likely a common source of compromise, where victims unknowingly installed a modified build of the tool alongside additional malicious components. The campaign then progressed into the next attack phase, marked by established connectivity to various C2 domains.

Beaconing Activity to "smshero"-themed domains

Darktrace subsequently observed the same infected device connecting to various C2 domains used to retrieve configuration data. As such, these external hostnames were themed around the string “smshero”, for example ‘smshero[.]co’.

On January 8th, Darktrace observed SSL beaconing to a rare destination which was attributed to a known ‘config/control domain’, nova[.]smshero[.]ai.
Figure 3: On January 8th, Darktrace observed SSL beaconing to a rare destination which was attributed to a known ‘config/control domain’, nova[.]smshero[.]ai.

The following day, on January 8, the device exhibited its first connectivity to a "smshero"-themed endpoint, which has since been identified as being associated with rotating C2 servers [1] [3]. Similar beaconing activity continued over the following days, with Darktrace identifying C2 connectivity to update[.]7zip[.]com over port 443, alongside additional connections to “smshero”‑themed endpoints such as zest.hero-sms[.]ai, flux.smshero[.]cc, and glide.smshero[.]cc between January 9 and January 15.

Darktrace later observed continued beaconing alerts over a 4-day interval to additional rare destinations attributed to a known ‘config/control domain’, zest[.]hero-sms[.]ai & glide[.]smshero[.]cc.
Figure 4: Darktrace later observed continued beaconing alerts over a 4-day interval to additional rare destinations attributed to a known ‘config/control domain’, zest[.]hero-sms[.]ai & glide[.]smshero[.]cc.

Proxied connectivity over destination ports

The primary objective of this campaign is believed to be proxyware, whereby third-party traffic is routed through victim devices to potentially obfuscate malicious activity. Devices were also observed communicating with rare external IPs hosted on Cloudflare and DataCamp Limited ASNs, establishing outbound proxy connections over the non-standard ports 1000 and 1002 [1].

OSINT sources also indicate that connections over these ports leveraged an XOR-encoded protocol (key 0x70) designed to obscure control messages. While the end goal of the campaign remains unclear, residential proxy networks can be abused to evade security rules and facilitate further unauthorized activities, including phishing and malware distribution [1][3].

Specifically, on January 8, Darktrace observed the device engaging in low-and-slow data exfiltration to the IP 79.127.221[.]47, which had first been observed the previous day, over port 1000. Proxyware typically installs an agent that routes third‑party traffic through an end-user’s device, effectively  turning it into a residential proxy exit node. This activity likely represents the system actively communicating outbound data to an entity that controls its behavior.

Figure 5: Darktrace later observed a ‘Low and Slow Exfiltration to IP’ alert, involving the known tunnel/proxy endpoint ‘79.127.221[.]47’.

Similar activity continued between January 10 and January 18, with Darktrace detecting threat actors attempting to exfiltrate significant volumes of data to 79.127.221[.]47 over destination port 1000.

Throughout the course of this incident, Darktrace’s Cyber AI Analyst launched several autonomous investigations, analyzing each anomalous event and ultimately painting a detailed picture of the attack timeline. These investigations correlated multiple incidents based on Darktrace detections observed between January 7 and January 19. Cyber AI Analyst identified anomalous variables such as repeated connections to unusual endpoints involving data uploads and downloads, with particular emphasis on HTTP and SSL connectivity.

Darktrace AI Analyst Coverage, showcasing multiple incident events that occurred on January 7th & 8th, highlighting associated malicious 7-zip behaviors.
Figure 6: Darktrace AI Analyst Coverage, showcasing multiple incident events that occurred on January 7th & 8th, highlighting associated malicious 7-zip behaviors.
Darktrace AI Analyst Endpoint Details from the given ‘Unusual Repeated Connections’ Incident Event, including the known tunnel/proxy endpoint.
Figure 7: Darktrace AI Analyst Endpoint Details from the given ‘Unusual Repeated Connections’ Incident Event, including the known tunnel/proxy endpoint.
Darktrace AI Analyst Coverage, showcasing additional incident events that occurred on January 12th through 18th, highlighting malicious 7-zip behaviors and SSL connectivity.
Figure 8: Darktrace AI Analyst Coverage, showcasing additional incident events that occurred on January 12th through 18th, highlighting malicious 7-zip behaviors and SSL connectivity.

Darktrace’s Autonomous Response

At several stages throughout the attack, Darktrace implemented Autonomous Response actions to help contain the suspicious activity as soon as it was identified, providing the customer’s security team with additional time to investigate and remediate. Between January 7 and January 18, Darktrace blocked a wide range of malicious activity, including beaconing connections to unusual endpoints, small data exfiltration attempts, and larger egress efforts, ultimately preventing the attacker from progressing through multiple stages of the attack or achieving their objectives.

Darktrace Autonomous Response Action Coverage showcasing connection block connection events including various endpoints that occurred on January 7th.
Figure 9: Darktrace Autonomous Response Action Coverage showcasing connection block connection events including various endpoints that occurred on January 7th.
Darktrace Antigena (Autonomous Response) Model Alert Coverage, showcasing a Antigena Suspicious Activity Block alert occurred on January 10th as a result of the Low and Slow Exfiltration to IP model alert.
Figure 10: Darktrace Antigena (Autonomous Response) Model Alert Coverage, showcasing a Antigena Suspicious Activity Block alert occurred on January 10th as a result of the Low and Slow Exfiltration to IP model alert.
Figure 11: Additional Darktrace Antigena (Autonomous Response) Model Alert Coverage, showcasing a Antigena Large Data Volume Outbound Block alert occurred on January 18th as a result of the Uncommon 1 GiB Outbound model alert.

Conclusion

The malicious 7‑Zip installer underscores how attackers continue to weaponize trust in widely used, legitimate software to gain initial access while evading user suspicion. By exploiting familiar and commonly installed services, this type of attack demonstrates that even routine actions, such as installing compression software, can become high‑risk events when defenses or user awareness are insufficient.

This campaign further emphasizes the urgent need for strict software validation and continuous network monitoring. Modern threats no longer rely solely on obscure tools or overtly malicious behavior. Instead, they increasingly blend seamlessly into everyday operations, making detection more challenging.

In this case, Darktrace / NETWORK was able to identify the anomalous activity and Autonomous Response actions in a timely manner, enabling the customer to be quickly notified and providing crucial additional time to investigate further.

In summary, the abuse of a trojanized 7‑Zip installer highlights a concerning shift in modern threat tactics, where trusted and widely deployed tools can serve as primary delivery mechanisms for system compromise. This reality reinforces that proactive detection, continuous monitoring, and strong security awareness are not optional but essential.

‍

Credit to Justin Torres, Senior Cyber Analyst, David Moreira da Silva, Cyber Analyst, Emma Foulger, Global Threat Research Operations Lead.

Edited by Ryan Traill (Content Manager)

Appendices

References

1. https://www.malwarebytes.com/blog/threat-intel/2026/02/fake-7-zip-downloads-are-turning-home-pcs-into-proxy-nodes

2. https://www.tomshardware.com/tech-industry/cyber-security/unofficial-7-zip-com-website-served-up-malware-for-10-days-files-turned-pcs-into-a-proxy-botnet

3. https://blog.lukeacha.com/2026/01/beware-of-fake-7zip-installer-upstage.html

4. https://www.bleepingcomputer.com/news/security/malicious-7-zip-site-distributes-installer-laced-with-proxy-tool/

5. https://customerportal.darktrace.com/guides/antigena-network-model-actions

Darktrace Model Detections

·      Anomalous Connection / Data Sent to Rare Domain

·      Anomalous Connection / Low and Slow Exfiltration to IP

·      Anomalous Connection / Multiple Failed Connections to Rare Endpoint

·      Anomalous Connection / Uncommon 1 GiB Outbound

·      Anomalous Server Activity / Rare External from Server

·      Compromise / Agent Beacon (Long Period)

·      Compromise / Beacon for 4 Days

·      Compromise / Beacon to Young Endpoint

·      Compromise / Beaconing Activity To External Rare

·      Compromise / High Volume of Connections with Beacon Score

·      Compromise / Large Number of Suspicious Failed Connections

·      Compromise / Large Number of Suspicious Successful Connections

·      Compromise / Repeating Connections Over 4 Days

·      Compromise / SSL Beaconing to Rare Destination

·      Compromise / Suspicious TLS Beaconing To Rare External

·      Device / Large Number of Model Alerts

·      Unusual Activity / Unusual External Activity

Cyber AI Analyst Coverage

·      Unusual Repeated Connections

·      Unusual Repeated Connections to Multiple Endpoints

·      Possible HTTP Command and Control

·      Possible HTTP Command and Control to Multiple Endpoints

·      Suspicious Remote Service Control Activity

·      Possible SSL Command and Control to Multiple Endpoints

Indicators of Compromise

IoC - Type - Description + Confidence

·      7zip[.]com – Hostname – C2 Endpoint

·      flux[.]smshero[.]co - Hostname - C2 Endpoint

·      neo[.]herosms[.]co - Hostname - C2 Endpoint

·      nova[.]smshero[.]ai - Hostname - C2 Endpoint

·      zest[.]hero-sms[.]ai -  Hostname - C2 Endpoint

·      soc[.]hero-sms[.]co - Hostname - C2 Endpoint

·      pulse[.]herosms[.]cc - Hostname - C2 Endpoint

·      glide[.]smshero[.]cc - Hostname - C2 Endpoint

·      prime[.]herosms[.]vip - Hostname - C2 Endpoint

·      172.96.115[.]226 - IP Address - C2 Endpoint

·      79.127.221[.]47:1002 – IP Address/Port - Proxy Endpoint

·      84.17.37[.]1:1002 - IP Address/Port - Proxy Endpoint

MITRE ATT&CK Mapping

Technique Name - Tactic - ID - Sub-Technique of

·      Exfiltration Over C2 Channel - EXFILTRATION - T1041

·      Scheduled Transfer - EXFILTRATION - T1029

·      Automated Exfiltration - EXFILTRATION - T1020

·      Data Transfer Size Limits - EXFILTRATION - T1030

·      External Proxy - COMMAND AND CONTROL - T1090.002 - T1090

·      Non-Application Layer Protocol - COMMAND AND CONTROL - T1095

·      Non-Standard Port - COMMAND AND CONTROL - T1571

·      Exfiltration to Cloud Storage - EXFILTRATION - T1567.002 - T1567

·      Exploit Public-Facing Application - INITIAL ACCESS - T1190

·      Web Protocols - COMMAND AND CONTROL - T1071.001 - T1071

·      Application Layer Protocol - COMMAND AND CONTROL - T1071

·      Man in the Browser - COLLECTION - T1185

·      Browser Extensions - PERSISTENCE - T1176

·      Encrypted Channel - COMMAND AND CONTROL - T1573

·      Fallback Channels - COMMAND AND CONTROL - T1008

·      Multi-Stage Channels - COMMAND AND CONTROL - T1104

·      Supply Chain Compromise - INITIAL ACCESS ICS - T0862

·      Commonly Used Port - COMMAND AND CONTROL ICS - T0885

Written by
Justin Torres
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
Justin Torres
Cyber Analyst

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

Network
•
June 24, 2026
Justin Torres
Cyber Analyst
Watch the NIS2 Webinar
Trending blogs
1
Prompt Security in Enterprise AI: Strengths, Weaknesses, and Common Approaches
May 20, 2026
2
How Empowering End Users can Improve Your Email Security and Decrease the Burden on the SOC
May 8, 2024
3
Elevating Network Security: Confronting Trust, Ransomware, & Novel Attacks
Jun 21, 2024
4
The State of AI in Cybersecurity: Unveiling Global Insights from 1,800 Security Practitioners
Apr 9, 2024
5
The State of AI in Cybersecurity: The Impact of AI on Cybersecurity Solutions
May 13, 2024

More in this series

No items found.
Continue reading
Network
•
June 24, 2026

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

Darktrace researchers analyze a recurring macOS intrusion pattern leveraging ClickFix-style social engineering and AppleScript post-compromise activity. By correlating weak behavioral signals such as HTTP POST beaconing, rare destinations, and SSL anomalies, defenders can detect and disrupt command-and-control activity before sustained attacker control is established.
Justin Torres
Cyber Analyst
Read more
Network
•
June 15, 2026

Hola VPN Abuse: From Proxy Traffic to Malware and Cryptomining

Darktrace’s analysis of Hola VPN-related activity across multiple environments reveals how peer-to-peer proxy functionality can enable malware delivery, command-and-control traffic, and cryptomining. Darktrace data highlights consistent patterns, suspicious downloads from rare endpoints, and post-compromise behavior, demonstrating how seemingly benign software can obscure threats and facilitate further malicious activity.
Min Kim
Cyber Security Analyst
Read more
Network
•
June 10, 2026

How Attackers Abuse the Chinese Nezha Monitoring Tool

This blog examines how attackers abuse the Chinese-developed Nezha monitoring tool, a legitimate open-source platform with remote access capabilities. Darktrace analysis of a honeypot intrusion highlights malicious deployment via containers, demonstrating how dual-use software enables stealthy operations, trust inversion, and detection challenges in dynamic cloud and enterprise environments modern.
Nathaniel Bill
Malware Research Engineer
Read more

Blog

/

Email

/

June 26, 2026

How Darktrace Transformed Cybersecurity at Our Health Center: A CIO’s Perspective

Default blog imageDefault blog image

How Darktrace Transformed Cybersecurity at Our Health Center: A CIO’s Perspective

In my role as CIO, I bring years of experience leading IT for healthcare organizations. I’ve seen firsthand the unique cybersecurity challenges that nonprofit health centers face: limited budgets, small IT teams, and the constant pressure to prioritize patient care over technology investments. Yet, the threat landscape for health is relentless, and the stakes for protecting patient data and ensuring operational continuity have never been higher. It’s a balancing act.

The search for a better solution

Like many nonprofits, organizations I work at start with Microsoft’s security stack. The discounted pricing for nonprofits makes it an obvious choice, and Microsoft Defender provided a solid foundation for endpoint and email security. However, I quickly realized that relying on a single vendor, even one as robust as Microsoft, left gaps in our defenses. Cybersecurity is never one-size-fits-all, which is why my preference was to layer an additional solution on top of our native security to improve our security posture.

Teams needed a solution that could layer seamlessly on top of Microsoft, without adding complexity or draining limited resources. That’s when I found Darktrace. I had heard of their reputation after seeing how other organizations used Darktrace to secure their infrastructure and was impressed by their AI-native, agentless approach and agreed to a proof of value (POV).

Our goal was to elavate Microsoft with an additional layer of intelligence- one that could seamlessly integrate, operate autonomously, and support a small team without increasing overhead. We turned to Darktrace because its AI-native, agentless approach offered a fundamentally different way to detect and respond to threats, learning our environment in real time and filling gaps that traditional tools can miss. With a quick POV, we were able to validate how effectively Darktrace works alongside Microsoft to deliver a more complete and resilient security architecture.

Why Darktrace stood out

From the start, Darktrace differentiated itself in several critical ways:

  • Deep visibility: Unlike other solutions that rely simply on host-based monitoring with endpoint agents, Darktrace operates passively at the network layer and integrates via APIs for email and identity security. This gave full visibility into network traffic that we previously didn’t have, going beyond our existing endpoint-based tools without adding additional maintenance overhead for our small IT team.
  • AI-native from the ground up: Darktrace wasn’t just layering AI on top of an existing product; it was built with AI at its core. Their autonomous detection and response to threats immediately reduced the need for constant human supervision. In a world where cyber-attacks are increasingly sophisticated and subtle, having an AI that learns our environment and adapts in real time is invaluable.
  • Comprehensive coverage: We started with a POV focused on email security, but quickly expanded to full deployment across our entire infrastructure. Darktrace’s products now protect our email, network, and identity layers, providing visibility and defense against lateral movement and abnormal behavior that traditional tools often miss.

Integration and workflow: Smooth and simple

One of the most impressive aspects of Darktrace is how easy it was to integrate into an existing environment. For network security, it was as simple as plugging an appliance into our top-of-rack switch – no downtime, no complex configuration. For email and identity, API integrations meant we could be up and running in hours, not weeks.

This simplicity extended to day-to-day operations. Our IT team received regular security reports, and any time we had questions or needed to adjust policies, Darktrace’s support team was there with white-glove service. Their responsiveness- even in the middle of the night- gave us confidence that we had true partners, not just a vendor.

Real-world impact: Threats stopped, time saved

The results spoke for themselves. During the time with Darktrace, I did not experience any security incidents. The team slept better at night knowing that Darktrace was monitoring for anomalies and proactively blocking suspicious activity, alerting us even before we noticed anything was wrong.

A memorable example was during an Electronic Health Record (EHR) upgrade, when my team forgot to adjust the policy in advance. Darktrace’s autonomous response was so effective that it blocked our upgrade activities- proof that nothing, not even internal changes, could slip by unnoticed. This level of vigilance meant that ransomware, data exfiltration attempts, or insider threats would be detected and contained before causing harm.

While I can’t share specific ROI numbers, the value was clear: we’ve avoided costly breaches, reduced the time spent investigating alerts, and eliminated the performance drag of agent-based tools. With Darktrace layered on top of Microsoft, I’ve hit the right balance of maximum protection with minimal spending. The cost of Darktrace / EMAIL was competitive, especially when factoring in the included Managed Detection and Response (MDR) service, which provides expert human oversight on top of the AI.

Key differentiators over the competition

  • Extending visibility beyond the endpoint: Traditional host-based monitoring solutions, such as EDR, play a critical role in securing individual devices. By adding a network detection and response (NDR) layer, we gained visibility into activity across our wider digital environment, surfacing threats that move laterally, operate between devices, or bypass endpoint controls. Darktrace also stood out for its ability to learn our normal patterns of behavior and identify subtle deviations in real time, not just known indicators of compromise. Because this is delivered through passive, non-disruptive monitoring, we were able to strengthen our defenses without adding complexity or impacting performance.
  • Layered security without complexity: Darktrace elevated our Microsoft foundation without creating conflicts or requiring us to disable existing protections. This layered approach maximized our security posture without adding operational burden.
  • Expert partnership: Beyond technology, Darktrace’s team acted as true partners, guiding us through deployment, providing ongoing support, and helping us interpret findings. This partnership was as valuable as the technology itself.

Advice for other nonprofits

If you’re an IT leader in a nonprofit, my advice is simple: look for solutions that are easy to deploy, intelligent in their response, and cost-effective. Don’t settle for more endpoint based tools that overlap with what you already have. Seek out a layered approach that covers your blind spots – especially at the network and email layers- at a price point that suits your organization.

Most importantly, don’t be afraid to evaluate new solutions. Even if you’re inundated with vendor pitches, you owe it to your organization to explore options that could save you time, money, and sleepless nights.

For organizations I work at, combining Microsoft’s security stack with Darktrace’s AI-native, platform struck the right balance between protection and practicality. We gained enterprise-grade security without sacrificing performance or stretching our budget. In the end, that meant more resources for what matters most: delivering care to our patients. If you’re facing similar challenges, I encourage you to consider how Darktrace could transform your security posture, and give your team the peace of mind they deserve.

For the organization I work in, combining Microsoft with Darktrace delivered a clear step-change in our security posture. Microsoft provided the foundation, while Darktrace’s behavioral intelligence added visibility into the unknown, surfacing emerging threats based on deviations in real-time activity, not just known indicators.

The result was enterprise-grade protection without added overhead, allowing us to stay focused on patient outcomes, not security operations. For organizations facing similar pressures, this layered approach offers a smarter, more efficient path to securing modern environments.

Continue reading
About the author
Mice Chen
Chief Information Security Officer

Blog

/

/

June 25, 2026

Shadow AI Detection: The First Step Toward Securing AI

shadow aiDefault blog imageDefault blog image

Why shadow AI is emerging  

Imagine you’re an employee under pressure, deadlines stacking up, repetitive tasks piling higher by the day. You find a free AI tool online that promises to automate the work in seconds; no approvals are needed. It feels like a simple win, paste in some data, write a quick prompt, and move faster.

But in that moment, something changed.  

Sensitive customer information is entered into a tool your organization doesn’t monitor, doesn’t govern, and can’t see and suddenly, that data is no longer where it should be, and no one knows where it’s gone.

This is the reality of Shadow AI: employees using unsanctioned AI tools to move faster, while unintentionally creating risk that exists entirely outside visibility and control.  

This is not just a one off case, research across businesses indicate that nearly half of employees report using unsanctioned AI tools, often prioritizing speed and productivity over security. Additionally, 51% of employees report connecting AI tools to work systems or apps without IT approval, creating significant operational risk where the average cost of security incidents in organizations with a high level of shadow AI usage can reach $670k.

While shadow AI is often top of mind for security professionals, it is just one component of how AI use can increase risk. Understanding and managing shadow AI use should be considered as part of a broader, comprehensive risk management strategy that aims to secure AI systems, including human and agent identities, interactions, human-AI partnerships, and behaviors operating across the digital enterprise from visibility and governance through detection, response, and recovery.  

Effective risk management calls for a layered and interdisciplinary strategy. It requires addressing issues across governance and visibility; identity, access and agent control, data security and privacy, secure MLOps / LLMOps, runtime security, behavior-based detection, autonomous response and recovery.  

This blog explores a specific governance and visibility use case linked to shadow AI and reveals the challenges it presents as well as the defensive strategies that security teams can adopt.

Why shadow AI is hard to detect  

When it comes to AI, what organizations can easily see does not always reflect the full scope of AI activity occurring within the tools, applications, and workflows used across an enterprise. As a result, organizations using traditional rule-based methods to flag unusual activity may struggle to distinguish unsanctioned AI usage from legitimate operational behavior, particularly as SaaS applications, APIs, and orchestration layers increasingly have AI embedded into normal business workflows. Identifying threats using previously observed intelligence or depending on hard to maintain allow and block lists does not provide a dynamic enough strategy to manage risk. Also, many organizations are focusing on identifying Shadow AI in their governed infrastructure, like gateways, endpoints, or SASE, which is foundational. But, organizations require visibility and Shadow AI detection across all networked infrastructure from on-prem, hybrid, data centers, and cloud infrastructure that may not have endpoint agent visibility. This uncovers the utilization of MCP, data flows, and autonomous agents across these domains.

For example, employees interact with AI assistants across approved SaaS platforms every day. However, browser extensions and other types of plug-ins can route prompts that include enterprise data to embedded AI services in ways that are not visible to the security team. AI enabled workflows may invoke multiple APIs, orchestration layers, and cloud services behind the scenes, making it difficult for traditional security tooling to determine where data is processed, stored, or retransmitted. Because much of this activity occurs within trusted browser sessions and encrypted SaaS traffic, conventional network monitoring, DLP, and application allowlisting controls often lack the context needed to accurately identify or govern these interactions

Identifying AI tools in the environment is one part of the equation. Understanding the behavior surrounding their use is where the real challenge lies. An AI application is not inherently risky, but the way users or other assets interact with it may be. Sensitive data exposure, abnormal access patterns, and misuse of AI-assisted workflows often appear legitimate in isolation and only become visible through behavioral analysis across the broader environment.  

What Shadow AI visibility does and doesn’t show

Comprehensive Shadow AI visibility allows organizations to answer several important questions:

  • What types of AI are we using? What AI platforms, agents, MCP clients/servers, and services are active across the enterprise?  
  • Who is using AI services? Which users, business units, or systems are interacting with those AI services?  
  • Is our data safe? Is sensitive or regulated data being exposed through prompts, workflows, or integrations?  
  • Are AI systems behaving as expected? Are AI systems behaving anomalously or operating outside approved governance processes?  
  • Are our AI systems under attack? Is an attacker attempting to manipulate prompts, influence agent behavior, or abuse AI-enabled workflows?

Answering these questions is foundational to broader AI governance efforts. However, it is limited to helping teams understand initial interactions and fails to offer insight into dependencies and outcomes that are critical to securing AI across an enterprise.  

Deeper visibility that includes the ability to understand dependencies and outcomes are not always available in AI security point products. Answering the questions below requires understanding runtime behavior and operational outcomes:  

  • What actions did the AI interaction trigger?  
  • What systems, applications, or data did it access? Did the AI operate beyond its intended permissions or scope?  
  • Could a low-risk interaction lead to high-risk outcomes?  
  • What is the risk and context understanding of an anomalous activity to assist in prioritization of analysis and autonomous response action?

The distinction between these two sets of questions offers two different layers of AI security. The first set of questions focuses on discovery and interaction visibility. The second set focuses on providing visibility that includes the context and outcomes that are critical for managing follow-on risks associated with obfuscated downstream activities.  

Together, these layers help organizations move beyond simply identifying AI usage toward understanding how AI behaves operationally across the enterprise.

How organizations are addressing shadow AI

Most organizations still approach shadow AI as an application control problem, relying on policies, browser restrictions, and allow/block lists. However, AI adoption is evolving faster than most governance processes can realistically keep pace with. New assistants, plugins, and embedded AI features appear continuously, creating pressure to enable business productivity while simultaneously containing risk.  

Existing governance processes were designed for a more traditional SaaS adoption cycle, where new applications could be reviewed, approved, and monitored over longer time horizons. AI adoption operates differently. New capabilities can appear overnight inside existing platforms employees already use, making it difficult for security and governance teams to maintain an accurate understanding of enterprise AI exposure. This means that many organizations are experiencing significant operational overhead, particularly in large environments where AI usage is decentralized across teams, departments, and third-party services.  

Where should organizations start when securing their AI systems?

Shadow AI identification is an on-going critical component for AI Risk/Governance Boards as well as security organizations. As organizations seek AI certifications like ISO 42001 AI Management Systems, visibility into all AI adoption from enterprise use to custom innovation and development is crucial. Shadow AI identification provides organizations with the visibility needed to decide whether an AI tool should be brought into governed environments to reduce data loss (DLP) risks or whether policies should be established and enforced to restrict their use.

As organizations rapidly innovate and adopt AI, they are taking on more and more risk. Organizations need to have a strategy in place to mitigate the assumed risk, especially with third-party adoption. Visibility, monitoring, governance enforcement, behavioral-based detection of non-deterministic systems, and autonomous investigation and containment becomes critical to mitigating the risk of AI systems.  

How Darktrace secures AI and shadow AI

Attackers are using AI to move faster, scale tactics, and make threats more adaptive and convincing. Internally, organizations are grappling with new forms of risk created by generative AI, autonomous agents, shadow AI, and increasingly complex digital environments.

Darktrace helps organizations protect both people and AI in a world where AI is now central to how business gets done. Darktrace / SECURE AI helps organizations discover and control shadow AI by surfacing unsanctioned or unexpected AI activity where it appears – including MCP detections, distinguishing misuse of legitimate tools and unapproved services, and applying policy to contain data exposure while guiding users toward sanctioned options.

Stay up to date on AI security

Sign up for the Secure AI Readiness Program here: This gives you exclusive access to the latest news on the latest AI threats, updates on emerging approaches shaping AI security, and insights into the latest innovations, including Darktrace’s ongoing work in this area.

Ready to talk with a Darktrace expert on securing AI? Register here to receive practical guidance on the AI risks that matter most to your business, paired with clarity on where to focus first across governance, visibility, risk reduction, and long-term readiness.  

Continue reading
About the author
Your data. Our AI.
Elevate your network security with Darktrace AI