Why Data Classification Isn’t Enough to Prevent Data Loss
In a world of growing data volume and diversity, protecting and keeping track of your organization’s sensitive information is increasingly complex – particularly when 63% of breaches stem from malicious insiders or human error. This blog explores how security teams can achieve visibility beyond the limits of data classification, without adding to the burden of data management.
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
Carlos Gray
Senior Product Marketing Manager, Email
Share
15
Apr 2025
Why today’s data is fundamentally difficult to protect
Data isn’t what it used to be. It’s no longer confined to neat rows in a database, or tucked away in a secure on-prem server. Today, sensitive information moves freely between cloud platforms, SaaS applications, endpoints, and a globally distributed workforce – often in real time. The sheer volume and diversity of modern data make it inherently harder to monitor, classify, and secure. And the numbers reflect this challenge – 63% of breaches stem from malicious insiders or human error.
This complexity is compounded by an outdated reliance on manual data management. While data classification remains critical – particularly to ensure compliance with regulations like GDPR or HIPAA – the burden of managing this data often falls on overstretched security teams. Security teams are expected to identify, label, and track data across sprawling ecosystems, which can be time-consuming and error-prone. Even with automation, rigid policies that depend on pre-defined data classification miss the mark.
From a data protection perspective, if manual or basic automated classification is the sole methodology for preventing data loss, critical data will likely slip through the cracks. Security teams are left scrambling to fill the gaps, facing compliance risks and increasing operational overhead. Over time, the hidden costs of these inefficiencies pile up, draining resources and reducing the effectiveness of your entire security posture.
What traditional data classification can’t cover
Data classification plays an important role in data loss prevention, but it's only half the puzzle. It’s designed to spot known patterns and apply labels, yet the most common causes of data breaches don’t follow rules. They stem from something far harder to define: human behavior.
Data classification is blind to nuance – it can’t grasp intent, context, or the subtle red flags that often precede a breach. And no amount of labeling, policy, or training can fully account for the reality that humans make mistakes. These problems require a system that sees beyond the data itself — one that understands how it’s being used, by whom, and in what context. That’s why Darktrace leans into its core strength: detecting the subtle symptoms of data loss by interpreting human behavior, not just file labels.
Achieving autonomous data protection with behavioral AI
Its understanding of business operations allows it to detect subtle anomalies around data movement for your use cases, whether that’s a misdirected email, an insecure cloud storage link, or suspicious activity from an insider. Crucially, this detection is entirely autonomous, with no need for predefined rules or static labels.
Fig 1: Darktrace uses its contextual understanding of each user to stop all types of sensitive or misdirected data from leaving the organization
Darktrace / EMAIL’s DLP add-on continuously learns in real time, enabling:
Automatic detection: Identifies risky data behavior to catch threats that traditional approaches miss – from human error to sophisticated insider threats.
A dynamic range of actions: Darktrace always aims to avoid business disruption in its blocking actions, but this can be adjusted according to the unique risk appetite of each customer – taking the most appropriate response for that business from a whole scale of possibilities.
Enhanced context: While Darktrace doesn’t require sensitivity data labeling, it integrates with Microsoft Purview to ingest sensitivity labels and enrich its understanding of the data – for even more accurate decision-making.
Beyond preventing data loss, Darktrace uses DLP activity to enhance its contextual understanding of the user itself. In other words, outbound activity can be a useful symptom in identifying a potential account compromise, or can be used to give context to that user’s inbound activity. Because Darktrace sees the whole picture of a user across their inbound, outbound, and lateral mail, as well as messaging (and into collaboration tools with Darktrace / IDENTITY), every interaction informs its continuous learning of normal.
With Darktrace, you can achieve dynamic data loss prevention for the most challenging human-related use cases – from accidental misdirected recipients to malicious insiders – that evade detection from manual classification. So don’t stand still on data protection – make the switch to autonomous, adaptive DLP that understands your business, data, and people.
[related-resource]
Interested in finding out more?
Read the full solution brief to see how Darktrace's AI-driven approach to DLP stops data loss across email and Teams
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.
Email-Borne Cyber Risk: A Core Challenge for the CISO in the Age of Volume and Sophistication
A former CISO shares his perspective on the challenge of securing the human layer, how existing security awareness training falls short, and how it can be improved to better prepare for high-volume, high-impact, human-centric threats.
Darktrace Unites Human Behavior and Threat Detection Across Email, Slack, Teams, and Zoom
Introducing the adaptive era of email security: a unified platform that connects personalized coaching, collaboration tools, and user behavior into a self-improving defense system.
Why Organizations are Moving to Label-free, Behavioral DLP for Outbound Email
Modern data loss doesn’t always look like a regex match. It can look like everyday communication slightly out of context. Here’s how a domain specific language model paired with behavioral learning protects labeled and unlabeled data without slowing business down.
Mythos vs Ethos: Defending in an Era of AI‑Accelerated Vulnerability Discovery
Anthropic’s Mythos and what it means for security teams
Recent attention on systems such as Anthropic Mythos highlights a notable problem for defenders. Namely that disclosure’s role in coordinating defensive action is eroding.
As AI systems gain stronger reasoning and coding capability, their usefulness in analyzing complex software environments and identifying weaknesses naturally increases. What has changed is not attacker motivation, but the conditions under which defenders learn about and organize around risk. Vulnerability discovery and exploitation increasingly unfold in ways that turn disclosure into a retrospective signal rather than a reliable starting point for defense.
Faster discovery was inevitable and is already visible
The acceleration of vulnerability discovery was already observable across the ecosystem. Publicly disclosed vulnerabilities (CVEs) have grown at double-digit rates for the past two years, including a 32% increase in 2024 according to NIST, driven in part by AI even prior to Anthropic’s Mythos model. Most notably XBOW topped the HackerOne US bug bounty leaderboard, marking the first time an autonomous penetration tester had done so.
The technical frontier for AI capabilities has been described elsewhere as jagged, and the implication is that Mythos is exceptional but not unique in this capability. While Mythos appears to make significant progress in complex vulnerability analysis, many other models are already able to find and exploit weaknesses to varying degrees.
What matters here is not which model performs best, but the fact that vulnerability discovery is no longer a scarce or tightly bounded capability.
The consequence of this shift is not simply earlier discovery. It is a change in the defender-attacker race condition. Disclosure once acted as a rough synchronization point. While attackers sometimes had earlier knowledge, disclosure generally marked the moment when risk became visible and defensive action could be broadly coordinated. Increasingly, that coordination will no longer exist. Exploitation may be underway well before a CVE is published, if it is published at all.
Why patch velocity alone is not the answer
The instinctive response to this shift is to focus on patching faster, but treating patch velocity as the primary solution misunderstands the problem. Most organizations are already constrained in how quickly they can remediate vulnerabilities. Asset sprawl, operational risk, testing requirements, uptime commitments, and unclear ownership all limit response speed, even when vulnerabilities are well understood.
If discovery and exploitation now routinely precede disclosure, then patching cannot be the first line of defense. It becomes one necessary control applied within a timeline that has already shifted. This does not imply that organizations should patch less. It means that patching cannot serve as the organizing principle for defense.
Defense needs a more stable anchor
If disclosure no longer defines when defense begins, then defense needs a reference point that does not depend on knowing the vulnerability in advance.
Every digital environment has a behavioral character. Systems authenticate, communicate, execute processes, and access resources in relatively consistent ways over time. These patterns are not static rules or signatures. They are learned behaviors that reflect how an organization operates.
When exploitation occurs, even via previously unknown vulnerabilities, those behavioral patterns change.
Attackers may use novel techniques, but they still need to gain access, create processes, move laterally, and will ultimately interact with systems in ways that diverge from what is expected. That deviation is observable regardless of whether the underlying weakness has been formally named.
In an environment where disclosure can no longer be relied on for timing or coordination, behavioral understanding is no longer an optional enhancement; it becomes the only consistently available defensive signal.
Detecting risk before disclosure
Darktrace’s threat research has consistently shown that malicious activity often becomes visible before public disclosure.
This reflects a defensive approach grounded in ‘Ethos’, in contrast to the unbounded exploration represented by ‘Mythos’. Here, Mythos describes continuous vulnerability discovery at speed and scale. Ethos reflects an understanding of what is normal and expected within a specific environment, grounded in observed behavior.
Revisiting assume breach
These conditions reinforce a principle long embedded in Zero Trust thinking: assume breach.
If exploitation can occur before disclosure, patching vulnerabilities can no longer act as the organizing principle for defense. Instead, effective defense must focus on monitoring for misuse and constraining attacker activity once access is achieved. Behavioral monitoring allows organizations to identify early‑stage compromise and respond while uncertainty remains, rather than waiting for formal verification.
AI plays a critical role here, not by predicting every exploit, but by continuously learning what normal looks like within a specific environment and identifying meaningful deviation at machine speed. Identifying that deviation enables defenders to respond by constraining activity back towards normal patterns of behavior.
Not an arms race, but an asymmetry
AI is often framed as fueling an arms race between attackers and defenders. In practice, the more important dynamic is asymmetry.
Attackers operate broadly, scanning many environments for opportunities. Defenders operate deeply within their own systems, and it’s this business context which is so significant. Behavioral understanding gives defenders a durable advantage. Attackers may automate discovery, but they cannot easily reproduce what belonging looks like inside a particular organization.
A changed defensive model
AI‑accelerated vulnerability discovery does not mean defenders have lost. It does mean that disclosure‑driven, patch‑centric models no longer provide a sufficient foundation for resilience.
As vulnerability volumes grow and exploitation timelines compress, effective defense increasingly depends on continuous behavioral understanding, detection that does not rely on prior disclosure, and rapid containment to limit impact. In this model, CVEs confirm risk rather than define when defense begins.
The industry has already seen this approach work in practice. As AI continues to reshape both offense and defense, behavioral detection will move from being complementary to being essential.
To observe adversary behavior in real time, Darktrace operates a global honeypot network known as “CloudyPots”, designed to capture malicious activity across a wide range of services, protocols, and cloud platforms. These honeypots provide valuable insights into the techniques, tools, and malware actively targeting internet‑facing infrastructure.
How attackers used a Jenkins honeypot to deploy the botnet
One such software honeypotted by Darktrace is Jenkins, a CI build system that allows developers to build code and run tests automatically. The instance of Jenkins in Darktrace’s honeypot is intentionally configured with a weak password, allowing attackers to obtain remote code execution on the service.
In one instance observed by Darktrace on March 18, 2026, a threat actor seemingly attempted to target Darktrace’s Jenkins honeypot to deploy a distributed denial-of-service (DDoS) botnet. Further analysis by Darktrace’s Threat Research team revealed the botnet was intended to specifically target video game servers.
How the Jenkins scriptText endpoint was used for remote code execution
The Jenkins build system features an endpoint named scriptText, which enables users to programmatically send new jobs, in the form of a Groovy script. Groovy is a programming language with similar syntax to Java and runs using the Java Virtual Machine (JVM). An attacker can abuse the scriptText endpoint to run a malicious script, achieving code execution on the victim host.
Figure 1: Request sent to the scriptText endpoint containing the malicious script.
The malicious script is sent using the form-data content type, which results in the contents of the script being URL encoded. This encoding can be decoded to recover the original script, as shown in Figure 2, where Darktrace Analysts decoded the script using CyberChef,
Figure 2: The malicious script decoded using CyberChef.
What happens after Jenkins is compromised
As Jenkins can be deployed on both Microsoft Windows and Linux systems, the script includes separate branches to target each platform.
In the case of Windows, the script performs the following actions:
Downloads a payload from 103[.]177.110.202/w.exe and saves it to C:\Windows\Temp\update.dat.
Renames the “update.dat” file to “win_sys.exe” (within the same folder)
Runs the Unblock-File command is used to remove security restrictions typically applied to files downloaded from the internet.
Adds a firewall allow rule is added for TCP port 5444, which the payload uses for command-and-control (C2) communications.
On Linux systems, the script will instead use a Bash one-liner to download the payload from 103[.]177.110.202/bot_x64.exe to /tmp/bot and execute it.
Why this botnet uses a single IP for delivery and command and control
The IP 103[.]177.110.202 belongs to Webico Company Limited, specifically its Tino brand, a Vietnamese company that offers domain registrar services and server hosting. Geolocation data indicates that the IP is located in Ho Chi Minh City. Open-source intelligence (OSINT) analysis revealed multiple malicious associations tied to the IP [1].
Darktrace’s analysis found that the IP 103[.]177.110.202 is used for multiple stages of an attack, including spreading and initial access, delivering payloads, and C2 communication. This is an unusual combination, as many malware families separate their spreading servers from their C2 infrastructure. Typically, malware distribution activity results in a high volume of abuse complaints, which may result in server takedowns or service suspension by internet providers. Separate C2 infrastructure ensures that existing infections remain controllable even if the spreading server is disrupted.
How the malware evades detection and maintains persistence
Analysis of the Linux payload (bot _x64)
The sample begins by setting the environmental variables BUILD_ID and JENKINS_NODE_COOKIE to “dontKillMe”. By default, Jenkins terminates long-running scripts after a defined timeout period; however, setting these variables to “dontKillMe” bypasses this check, allowing the script to continue running uninterrupted.
The script then performs several stealth behaviors to evade detection. First, it deletes the original executable from disk and then renames itself to resemble the legitimate kernel processes “ksoftirqd/0” or “kworker”, which are found on Linux installations by default. It then uses a double fork to daemonize itself, enabling it to run in the background, before redirecting standard input, standard output, and standard error to /dev/null, hiding any logging from the malware. Finally, the script creates a signal handler for signals such as SIGTERM, causing them to be ignored and making it harder to stop the process.
Figure 3: Stealth component of the main function
How the botnet communicates with command and control (C2)
The sample then connects to the C2 server and sends the detected architecture of the system on which the agent was installed. The malware then enters a loop to handle incoming commands.
The sample features two types of commands, utility commands used to manage the malware, and commands to trigger attacks. Three special commands are defined: “PING” (which replies with PONG as a keep-alive mechanism), “!stop” which causes the malware to exit, and “!update”, which triggers the malware to download a new version from the C2 server and restart itself.
Figure 4: Initial connection to the C2 sever.
What DDoS attack techniques this botnet uses
The attack commands consist of the following:
Many of these commands invoke the same function despite appearing to be different attack techniques. For example, specialized attacks such as Cloudflare bypass (cfbypass, uam) use the exact same function as a standard HTTP attack. This may indicate the threat actor is attempting to make the botnet look like it has more capabilities than it actually has, or it could suggest that these commands are placeholders for future attack functionality that has yet to be implemented
All the commands take three arguments: IP, port to attack, and the duration of the attack.
attack_udp and attack_udp_pps
The attack_udp and attack_udp_pps functions both use a basic loop and sendto system call to send UDP packets to the victim’s IP, either targeting a predetermined port or a random port. The attack_udp function sends packets with 1,450 bytes of data, aimed at bandwidth saturation, while the attack_udp_pps function sends smaller 64-byte packets. In both cases, the data body of the packet consists of entirely random data.
Figure 5: Code for the UDP attack method
attack_dayz
The attack_dayz function follows a similar structure to the attack_udp function; however, instead of sending random data, it will instead send a TSource Engine Query. This command is specific to Valve Source Engine servers and is designed to return a large volume of data about the targeted server. By repeatedly flooding this request, an attacker can exhaust the resources of a server using a comparatively small amount of data.
The Valve Source Engine server, also called Source Engine Dedicated server, is a server developed by video game company Valve that enables multiplayer gameplay for titles built using the Source game engine, which is also developed by Valve. The Source engine is used in games such as Counterstrike and Team Fortress 2. Curiously, the function attack_dayz, appears to be named after another popular online multiplayer game, DayZ; however, DayZ does not use the Valve Source Engine, making it unclear why this name was chosen.
Figure 6: The code for the “attack_dayz” attack function.
attack_tcp_push
The attack_tcp_push function establishes a TCP socket with the non-blocking flag set, allowing it to rapidly call functions such as connect() and send() without waiting for their completion. For the duration of the attack, it enters a while loop in which it repeatedly connects to the victim, sends 1,024 bytes of random data, and then closes the connection. This process repeats until the attack duration ends. If the mode flag is set to 1, the function also configures the socket with TCP no-delay enabled, allowing for packets to be sent immediately without buffering, resulting in a higher packet rate and a more effective attack.
Figure 7: The code for the TCP attack function.
attack_http
Similar to attach_tcp_push, attack_http configures a socket with no-delay enabled and non-blocking set. After establishing the connection, it sends 64 HTTP GET requests before closing the socket.
Figure 8: The code for the HTTP attack function.
attack_special
The attack_special function creates a UDP socket and sets the port and payload based on the value of the mode flag:
Mode 0: Port 53 (DNS), sending a 10-byte malformed data packet.
Mode 1: Port 27015 (Valve Source Engine), sending the previously observed TSource Engine Query packet.
Mode 2: Port 123 (NTP), sending the start of an NTP control request.
Figure 9: The code for the attack_special function.
What this botnet reveals about opportunistic attacks on internet-facing systems
Jenkins is one of the less frequently exploited services honeypotted by Darktrace, with only a handful campaigns observed. Nonetheless, the emergence of this new DDoS botnet demonstrates that attackers continue to opportunistically exploit any internet-facing misconfiguration at scale to grow the botnet strength.
While the hosts most commonly affected by these opportunistic attacks are usually “lower-value” systems, this distinction is largely irrelevant for botnets, where numbers alone are more important to overall effectiveness
The presence of game-specific DoS techniques further highlights that the gaming industry continues to be extensively targeted by cyber attackers, with Cloudflare reporting it as the fourth most targeted industry [2]. This botnet has likely already been used against game servers, serving as a reminder for server operators to ensure appropriate mitigations are in place.
Credit to Nathaniel Bill (Malware Research Engineer) Edited by Ryan Traill (Content Manager)
Indicators of Compromise (IoCs)
103[.]177.110.202 - Attacker and command-and-control IP
.avif)
.avif)
