Blog
/
AI
/
April 16, 2025

Introducing Version 2 of Darktrace’s Embedding Model for Investigation of Security Threats (DEMIST-2)

Learn how Darktrace’s DEMIST-2 embedding model delivers high-accuracy threat classification and detection across any environment, outperforming larger models with efficiency and precision.
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
Margaret Cunningham, PhD
VP, Security & AI Strategy, Field CISO
woman looking at laptop at deskDefault blog image
16
Apr 2025

DEMIST-2 is Darktrace’s latest embedding model, built to interpret and classify security data with precision. It performs highly specialized tasks and can be deployed in any environment. Unlike generative language models, DEMIST-2 focuses on providing reliable, high-accuracy detections for critical security use cases.

DEMIST-2 Core Capabilities:  

  • Enhances Cyber AI Analyst’s ability to triage and reason about security incidents by providing expert representation and classification of security data, and as a part of our broader multi-layered AI system
  • Classifies and interprets security data, in contrast to language models that generate unpredictable open-ended text responses  
  • Incorporates new innovations in language model development and architecture, optimized specifically for cybersecurity applications
  • Deployable across cloud, on-prem, and edge environments, DEMIST-2 delivers low-latency, high-accuracy results wherever it runs. It enables inference anywhere.

Cybersecurity is constantly evolving, but the need to build precise and reliable detections remains constant in the face of new and emerging threats. Darktrace’s Embedding Model for Investigation of Security Threats (DEMIST-2) addresses these critical needs and is designed to create stable, high-fidelity representations of security data while also serving as a powerful classifier. For security teams, this means faster, more accurate threat detection with reduced manual investigation. DEMIST-2's efficiency also reduces the need to invest in massive computational resources, enabling effective protection at scale without added complexity.  

As an embedding language model, DEMIST-2 classifies and creates meaning out of complex security data. This equips our Self-Learning AI with the insights to compare, correlate, and reason with consistency and precision. Classifications and embeddings power core capabilities across our products where accuracy is not optional, as a part of our multi-layered approach to AI architecture.

Perhaps most importantly, DEMIST-2 features a compact architecture that delivers analyst-level insights while meeting diverse deployment needs across cloud, on-prem, and edge environments. Trained on a mixture of general and domain-specific data and designed to support task specialization, DEMIST-2 provides privacy-preserving inference anywhere, while outperforming larger general-purpose models in key cybersecurity tasks.

This proprietary language model reflects Darktrace's ongoing commitment to continually innovate our AI solutions to meet the unique challenges of the security industry. We approach AI differently, integrating diverse insights to solve complex cybersecurity problems. DEMIST-2 shows that a refined, optimized, domain-specific language model can deliver outsized results in an efficient package. We are redefining possibilities for cybersecurity, but our methods transfer readily to other domains. We are eager to share our findings to accelerate innovation in the field.  

The evolution of DEMIST-2

Key concepts:  

  • Tokens: The smallest units processed by language models. Text is split into fragments based on frequency patterns allowing models to handle unfamiliar words efficiently
  • Low-Rank Adaptors (LoRA): Small, trainable components added to a model that allow it to specialize in new tasks without retraining the full system. These components learn task-specific behavior while the original foundation model remains unchanged. This approach enables multiple specializations to coexist, and work simultaneously, without drastically increasing processing and memory requirements.

Darktrace began using large language models in our products in 2022. DEMIST-2 reflects significant advancements in our continuous experimentation and adoption of innovations in the field to address the unique needs of the security industry.  

It is important to note that Darktrace uses a range of language models throughout its products, but each one is chosen for the task at hand. Many others in the artificial intelligence (AI) industry are focused on broad application of large language models (LLMs) for open-ended text generation tasks. Our research shows that using LLMs for classification and embedding offers better, more reliable, results for core security use cases. We’ve found that using LLMs for open-ended outputs can introduce uncertainty through inaccurate and unreliable responses, which is detrimental for environments where precision matters. Generative AI should not be applied to use cases, such as investigation and threat detection, where the results can deeply matter. Thoughtful application of generative AI capabilities, such as drafting decoy phishing emails or crafting non-consequential summaries are helpful but still require careful oversight.

Data is perhaps the most important factor for building language models. The data used to train DEMIST-2 balanced the need for general language understanding with security expertise. We used both publicly available and proprietary datasets.  Our proprietary dataset included privacy-preserving data such as URIs observed in customer alerts, anonymized at source to remove PII and gathered via the Call Home and aianalyst.darktrace.com services. For additional details, read our Technical Paper.  

DEMIST-2 is our way of addressing the unique challenges posed by security data. It recognizes that security data follows its own patterns that are distinct from natural language. For example, hostnames, HTTP headers, and certificate fields often appear in predictable ways, but not necessarily in a way that mirrors natural language. General-purpose LLMs tend to break down when used in these types of highly specialized domains. They struggle to interpret structure and context, fragmenting important patterns during tokenization in ways that can have a negative impact on performance.  

DEMIST-2 was built to understand the language and structure of security data using a custom tokenizer built around a security-specific vocabulary of over 16,000 words. This tokenizer allows the model to process inputs more accurately like encoded payloads, file paths, subdomain chains, and command-line arguments. These types of data are often misinterpreted by general-purpose models.  

When the tokenizer encounters unfamiliar or irregular input, it breaks the data into smaller pieces so it can still be processed. The ability to fall back to individual bytes is critical in cybersecurity contexts where novel or obfuscated content is common. This approach combines precision with flexibility, supporting specialized understanding with resilience in the face of unpredictable data.  

Along with our custom tokenizer, we made changes to support task specialization without increasing model size. To do this, DEMIST-2 uses LoRA . LoRA is a technique that integrates lightweight components with the base model to allow it to perform specific tasks while keeping memory requirements low. By using LoRA, our proprietary representation of security knowledge can be shared and reused as a starting point for more highly specialized models, for example, it takes a different type of specialization to understand hostnames versus to understand sensitive filenames. DEMIST-2 dynamically adapts to these needs and performs them with purpose.  

The result is that DEMIST-2 is like having a room of specialists working on difficult problems together, while sharing a basic core set of knowledge that does not need to be repeated or reintroduced to every situation. Sharing a consistent base model also improves its maintainability and allows efficient deployment across diverse environments without compromising speed or accuracy.  

Tokenization and task specialization represent only a portion of the updates we have made to our embedding model. In conjunction with the changes described above, DEMIST-2 integrates several updated modeling techniques that reduce latency and improve detections. To learn more about these details, our training data and methods, and a full write-up of our results, please read our scientific whitepaper.

DEMIST-2 in action

In this section, we highlight DEMIST-2's embeddings and performance. First, we show a visualization of how DEMIST-2 classifies and interprets hostnames, and second, we present its performance in a hostname classification task in comparison to other language models.  

Embeddings can often feel abstract, so let’s make them real. Figure 1 below is a 2D visualization of how DEMIST-2 classifies and understands hostnames. In reality, these hostnames exist across many more dimensions, capturing details like their relationships with other hostnames, usage patterns, and contextual data. The colors and positions in the diagram represent a simplified view of how DEMIST-2 organizes and interprets these hostnames, providing insights into their meaning and connections. Just like an experienced human analyst can quickly identify and group hostnames based on patterns and context, DEMIST-2 does the same at scale.  

DEMIST-2 visualization of hostname relationships from a large web dataset.
Figure 1: DEMIST-2 visualization of hostname relationships from a large web dataset.

Next, let’s zoom in on two distinct clusters that DEMIST-2 recognizes. One cluster represents small businesses (Figure 2) and the other, Russian and Polish sites with similar numerical formats (Figure 3). These clusters demonstrate how DEMIST-2 can identify specific groupings based on real-world attributes such as regional patterns in website structures, common formats used by small businesses, and other properties such as its understanding of how websites relate to each other on the internet.

Cluster of small businesses
Figure 2: Cluster of small businesses
Figure 3: Cluster of Russian and Polish sites with a similar numerical format

The previous figures provided a view of how DEMIST-2 works. Figure 4 highlights DEMIST-2’s performance in a security-related classification task. The chart shows how DEMIST-2, with just 95 million parameters, achieves nearly 94% accuracy—making it the highest-performing model in the chart, despite being the smallest. In comparison, the larger model with 278 million parameters achieves only about 89% accuracy, showing that size doesn’t always mean better performance. Small models don’t mean poor performance. For many security-related tasks, DEMIST-2 outperforms much larger models.

Hostname classification task performance comparison against comparable open source foundation models
Figure 4: Hostname classification task performance comparison against comparable open source foundation models

With these examples of DEMIST-2 in action, we’ve shown how it excels in embedding and classifying security data while delivering high performance on specialized security tasks.  

The DEMIST-2 advantage

DEMIST-2 was built for precision and reliability. Our primary goal was to create a high-performance model capable of tackling complex cybersecurity tasks. Optimizing for efficiency and scalability came second, but it is a natural outcome of our commitment to building a strong, effective solution that is available to security teams working across diverse environments. It is an enormous benefit that DEMIST-2 is orders of magnitude smaller than many general-purpose models. However, and much more importantly, it significantly outperforms models in its capabilities and accuracy on security tasks.  

Finding a product that fits into an environment’s unique constraints used to mean that some teams had to settle for less powerful or less performant products. With DEMIST-2, data can remain local to the environment, is entirely separate from the data of other customers, and can even operate in environments without network connectivity. The size of our model allows for flexible deployment options while at the same time providing measurable performance advantages for security-related tasks.  

As security threats continue to evolve, we believe that purpose-built AI systems like DEMIST-2 will be essential tools for defenders, combining the power of modern language modeling with the specificity and reliability that builds trust and partnership between security practitioners and AI systems.

Conclusion

DEMIST-2 has additional architectural and deployment updates that improve performance and stability. These innovations contribute to our ability to minimize model size and memory constraints and reflect our dedication to meeting the data handling and privacy needs of security environments. In addition, these choices reflect our dedication to responsible AI practices.

DEMIST-2 is available in Darktrace 6.3, along with a new DIGEST model that uses GNNs and RNNs to score and prioritize threats with expert-level precision.

[related-resource]

Want more details?

Read the full research paper to explore how DEMIST-2 was built, trained, and optimized to meet the unique challenges of cybersecurity

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
Margaret Cunningham, PhD
VP, Security & AI Strategy, Field CISO

More in this series

No items found.

Blog

/

AI

/

July 31, 2026

Building Operational Resilience Across Mission-Critical Marine Services

Default blog imageDefault blog image

Mission-Critical Marine Services

This marine organization supports offshore energy production, export infrastructure, and regional logistics, delivering critical services through its diverse fleet and a regional shorebase footprint. To power rapid mobilization and 24/7 operational readiness, the organization has embraced cloud adoption and digital transformation, reshaping how crews, contractors, and shore-based teams access services.

While technology modernization has enhanced operations, it has also introduced new security complexities.

  • From perimeter to identity: As access becomes more distributed, identity has become the primary security control, elevating the risk of credential compromise and privilege misuse.
  • From confidentiality to availability and resilience: As cloud platforms increasingly underpin fleet and operational systems, cyber incidents can disrupt services and safety.
  • From isolated tools to unified visibility:  Because Information Technology (IT) and Operational Technology (OT) often intersect, and legacy systems coexist with modern cloud platforms, fragmented monitoring makes it harder to understand risk and respond decisively across domains.
“Our cybersecurity priorities expanded along with our business goals, placing availability and resilience at the forefront. The impact of a potential threat became an operational risk, which elevated cybersecurity from an IT issue to an operational safety and resilience enabler.” - Information and Communications Technology (ICT) Manager.

A Unified, AI-Driven Platform for IT and OT

To strengthen visibility, detection, and response across its highly distributed environment, the customer adopted the Darktrace ActiveAI Security Platform™ in 2022.

Darktrace’s contextual detection capability was a key driver. Unlike traditional tools that rely on known threat signatures, Darktrace’s Self-Learning AI learns “normal” behavior to identify emerging threats and correlate visibility across the customer's siloed on-premises and cloud domains.

Today, the customer relies on:

  • Darktrace / EMAIL™ to reduce phishing risk and minimize disruption from legacy mail controls and false positives
  • Darktrace / IDENTITY™ to support identity-centric security as cloud access expands across vessels and shore-based operations
  • Darktrace / NETWORK™ to strengthen oversight across the broader environment, including operational contexts where IT and OT intersect
  • Darktrace / CLOUD™ (Azure), added in 2024, to extend detection and response into Azure and support cloud transformation without treating cloud as a separate security silo
  • Darktrace / Incident Readiness & Recovery to strengthen incident readiness and recovery planning
  • Darktrace Managed Detection and Response Services to provide 24/7/365 monitoring and support

This combination supports a single operating model for the customer: security that can adapt as the environment changes while remaining practical for a lean team responsible for safeguarding both business operations and safety-critical services.

Extending cloud protection without complexity

As the customer accelerated cloud adoption, it expanded coverage in 2024 with Darktrace / CLOUD for Azure to bring cloud workloads under the same AI-driven visibility and response model – without adding operational burden. “This matters in hybrid environments because attacks rarely stay in one place,” explains the ICT Manager. “A compromised identity can trigger activity in the cloud, which can open pathways back into on-premises systems.”

In parallel, Darktrace / CLOUD’s posture management capabilities support governance and audit readiness by surfacing misconfigurations and exposure risks earlier, before they become incidents.

A Stronger, Faster, More Resilient Business

Since adopting Darktrace, the customer has strengthened cyber resilience while reducing operational burden on its small ICT team.

Darktrace continuously analyzes millions of individual events that can contribute to a wider incident. Within a single month, the solution autonomously investigated 88% of all potential threats, taking appropriate action within just 39.4 seconds on average.

Autonomous capabilities ensure threats are stopped and contained until the ICT team can investigate. In one standout instance, Darktrace autonomously blocked malicious links during a mass phishing/spam event before other controls flagged the threat. the ICT Manager later confirmed Microsoft reported the link as malicious, but Darktrace had already acted to prevent delivery and reduce exposure.

“Whether something happens during off hours, while we’re on vacation, or when our attention is focused elsewhere, we’re confident Darktrace will take control and stop a threat before it spreads,” says the ICT Manager.

Darktrace’s Self-Learning AI combines multiple AI methods and advanced techniques to improve threat detection, investigation, and response dramatically reducing alert overload and manual triage. Within a single month, the solution saved the customer's IT group 411 equivalent human investigation hours.

“For a lean team supporting a 24/7 operational footprint, this autonomous action eliminates the constant firefighting and stress, giving us the space to focus on higher-value priorities.”- Information and Communications Technology (ICT) Manager.

Protecting communications without disruption

the customer experienced friction from legacy email and network controls prior to Darktrace, which generated high false positive rates, disrupted legitimate communications, created operational drag, and added workload for ICT. With Darktrace / EMAIL learning normal email behavior and applying context-aware actions, the team reduced unnecessary interruptions while maintaining protection.

“That shift matters in marine services, where business communications directly support coordination across vessels, shore bases, clients, ports, and regulators,” says the ICT Manager. “Darktrace doesn’t just block more threats, it autonomously makes decisions that preserve operational continuity and enable my team to focus on credible threats instead of chasing volume.”

Delivering clarity and confidence

Darktrace has reduced manual triage by correlating activity across email, identity, network, and cloud, providing the context needed to prioritize what matters without requiring the ICT team to stitch together evidence across multiple tools.

“With unified visibility we can identify patterns across domains, make informed decisions about where risk actually exists, and align security actions with operational impact rather than theoretical threats,” explains the ICT Manager. “I can now prioritize effort and investment across our ICT landscape with far greater confidence.”

Regular Executive Threat Reports reinforce operational confidence by giving leadership clear visibility into threats Darktrace has handled autonomously, supporting decisive action when needed and confidence to avoid unnecessary disruption when it isn’t.

Scaling Securely in a Hybrid World

As the customer advances its cloud transformation, the ICT Manager sees the Darktrace partnership evolving into a foundational layer of resilience and assurance, supporting scale, governance, and operational confidence in an increasingly cloud-centric environment.

Key priorities include:

  • Shifting from hybrid visibility to cloud-first resilience, using continuous monitoring and posture insights to reduce exposure earlier
  • Strengthening governance and audit readiness, especially as critical workloads and sensitive data expand in Azure and expectations rise under regulatory and client assurance requirements
  • Increasing reliance on autonomous response and AI investigation as the number of identities, workloads, and access paths grows faster than headcount
  • Deepening cross-domain correlation so cloud signals further enrich decision-making, supporting faster containment and more confident prioritization

“As we accelerate our cloud strategy, Darktrace will play an even more strategic role,” says the ICT Manager, “providing the guidance, technology, and expertise that allow us to grow with confidence and innovate securely.”

Continue reading
About the author

Blog

/

Network

/

July 28, 2026

Hiding in Plain Sight: Uncovering a Multi-Stage Ransomware Attack Through Behavioral Detection

ransomware attackDefault blog imageDefault blog image

Why ransomware has changed

Ransomware attacks have continued to increase globally, with 698 incidents reported in May 2026, representing a 48% rise compared to 472 incidents in May 2025 [1]. At the same time, the ransomware landscape is evolving. Several major ransomware groups, including LockBit [2], have been disrupted by successful joint law enforcement operations, resulting in a shift away from a small number of dominant actors towards a more fragmented and distributed ecosystem. This is increasingly composed of smaller groups who play a specialized role in the attack, such as initial access brokers, affiliates or developers.

As a result, ransomware tactics, techniques, and procedures (TTPs) are becoming more diverse and less predictable. On top of this, adversaries are leveraging native tools and legitimate penetration testing frameworks to evade detection. Anomaly-based detection is therefore critical to identify pre-ransomware activity, rather than relying on signatures associated with a handful of well-known ransomware groups.

As these attacks often unfold over several days, there is a critical window for defenders to act. In this context, behavioral-based detection plays a vital role in identifying suspicious pre-ransomware activity, and enabling early intervention before encryption or exfiltration occurs.

Inside a modern ransomware intrusion

In early 2026, Darktrace detected activity within a customer’s environment related to a multi-stage ransomware intrusion from the initial compromise. This activity does not appear to be attributable to a specific ransomware group, and no known ransomware payload was observed until the final stage.

The attack aligns with a broader industry trend in which compromised virtual private network (VPN) credentials are used as an entry point, followed by rapid internal reconnaissance and lateral movement using legitimate administrative tools. This growing preference for native tools and legitimate frameworks in cyber-attacks illustrates that it is increasingly unreliable to depend solely on traditional indicators of compromise such as known malware signatures or exploit detection.

The intrusion also involved the use of Sliver, an open-source adversary emulation framework, which is increasingly observed in real-world attacks. Originally designed for penetration testing and red teaming, Sliver has gained traction among threat actors as a stealthier alternative to more heavily signatured frameworks such as Cobalt Strike. As a legitimate framework, its use further complicates detection for security tools that rely on known malicious signatures.

Darktrace’s detection of a ransomware event in a customer’s environment

The initial compromise appears to have occurred via compromised credentials used over the VPN shortly before, or at the onset of the first indicators of suspicious activity. While it remains unclear as to how or when the threat actors gained access to these credentials, the use of initial access brokers (IABs) is a common feature of modern ransomware operations. This suggests that access to the environment may have been established several days or weeks beforehand.

The intrusion unfolded over three days, presenting multiple opportunities for early detection and intervention before ransomware deployment. The attack progressed through a compressed but structured sequence: initial access and reconnaissance were completed within hours, followed by privilege escalation and lateral movement the next day, and culminating in data exfiltration and encryption shortly thereafter. Throughout each stage, distinct behavioral anomalies emerged across the network providing clear indicators of malicious activity well before the ransomware was deployed.

While Darktrace’s Autonomous Response capability was enabled within the customer’s environment, it was not fully configured across the impacted devices, allowing the attack to progress to ransomware deployment. Had Autonomous Response been fully deployed across the affected systems, it could have taken targeted action against the earliest stages of malicious activity, potentially disrupting the intrusion before it escalated.

Figure 1: Timeline of the attack progression.

Day 1: Reconnaissance and privilege escalation

The threat actor gained access via compromised VPN credentials and initiated internal reconnaissance. Darktrace detected anomalous scanning behavior, including unusual port scanning activity and widespread network enumeration.

Specifically, Darktrace detected a high volume of east-west scanning activity across a broad range of ports, with TCP connections targeting ports 21, 80, 445, 4899 and 8080. Associated URIs suggested the use of Nmap, a widely used penetration testing tool. This highlights how attackers often leverage legitimate penetration testing tools for malicious reconnaissance, enabling them to blend into normal network activity and evade traditional signature-based detection methods.

Figure 2: Darktrace's detection of a sharp increase in anomalous internal connections, triggering multiple high-severity model alerts associated with reconnaissance activity.

Several devices were observed using administrative credentials to carry out privileged actions in a manner that was highly anomalous for the environment. This activity was accompanied by behavior consistent with SMB authentication scanning, suggesting efforts to identify and access additional systems. As the activity intensified, an increasing number of devices became involved, signalling lateral movement and further spread across the network.

Darktrace also identified privilege escalation through active directory (AD) replication abuse, specifically via the drsuapi::DRSGetNCChanges function. This technique allows an attacker with sufficient privileges to request directory replication data from a domain controller (DC), enabling them to extract credentials, including password hashes, without directly interacting with user accounts. Commonly associated with ‘DCSync’ attacks, this technique is frequently used to obtain highly privileged credentials and enable further escalation within an environment.

Figure 3: Darktrace’s detection of anomalous AD replication activity indicative of privilege escalation.

This activity was seen alongside the use of the now obsolete SMBv1, repeated NTLM authentication attempts using multiple variations of ‘Administrator’ credentials, reverse DNS scanning, and large-scale network scanning. Darktrace observed widespread use of SMBv1 across the customer’s environment, exposing a significant security weakness. As a legacy protocol with well-documented weaknesses, SMBv1 can be exploited to facilitate lateral movement, allowing the attackers to expand their access following initial compromise.

Day 3: Lateral Movement, Command & Control, and Exfiltration

Two days later, the attacker escalated privileges and expanded their foothold using living-off-the-land (LOTL) techniques such as PSExec, WMI, and RDP. Concurrently, Darktrace identified command-and-control (C2)-style communications consistent with the Sliver framework, alongside rare outbound connections to cloud infrastructure indicating potential data exfiltration. The volume and severity of observed activity increased as attack behavior intensified.

The device was observed conducting extensive lateral movement, leveraging LOTL techniques to evade detection. Activity included WMI execution (e.g. ExecQuery), DCE-RPC activity, SMB sessions and file writes, most of which were successful, as well as the deployment of PSEXESVC.exe via ADMIN$ shares and prolonged RDP sessions. Darktrace identified this behavior as highly anomalous for the environment. Such activity is commonly associated with the transfer of attacker tooling, remote command execution, and the establishment of persistent access across compromised systems.  

Figure 4: Darktrace’s detection of a spike in RPC binding events indicative of potential lateral movement.

On the same day, Darktrace detected C2-style SSL communications originating from multiple internal devices to rare external endpoints. These connections exhibited anomalous characteristics, including invalid SSL certificates and repeated connection patterns resembling beaconing. Analysis of the observed JA3 fingerprint further linked the activity to Sliver, the adversary simulation framework referenced earlier, as the hash has previously been associated with Sliver-related infrastructure [3]. The use of this framework reflects a broader trend of attackers repurposing legitimate offensive security tools for stealthy C2 communications. Connections to 137[.]220[.]59[.]55 (ASN AS20473 AS-VULTR) indicated that the communications were likely routed via a virtual private server (VPS) hosted by Vultr. Attackers often utilize VPS infrastructure from legitimate cloud providers like Vultr to obscure their true origin, blend into benign traffic, and evade IP-based detection mechanisms [4].

Figure 5: Darktrace’s Cyber AI Analyst detection of two linked unusual connections to Vultr infrastructure.

Darktrace also observed a device initiating SSL connections to safedata.s3[.]wasabisys[.]com, an endpoint associated with Wasabi cloud storage. Darktrace recognized that neither the destination nor the associated IP address had previously been observed within the environment.  More than 200 MB of data was subsequently uploaded externally to endpoints sharing the same JA3 client hash, indicating a sustained transfer session and potential data exfiltration to third-party storage. The apparent exfiltration prior to encryption is consistent with a double-extortion ransomware strategy.

Figure 6: Darktrace’s Cyber AI Analyst detection of more than 30 rare outbound connections to a Wasabi cloud storage endpoint, indicative of potential data exfiltration

Day 4: Encryption

The attack culminated in ransomware deployment, marking the transition from suspicious network activity to a business-impacting incident. Using SMB-based propagation, the threat actor encrypted thousands of files across the network, affecting multiple systems and disrupting normal operations. Throughout the encryption event, the legacy SMBv1 protocol was used extensively across multiple internal systems, resulting in a significant increase in newly encrypted files.

Figure 7: Darktrace’s detection of abnormal spikes in SMB activity and associated model alerts, indicative of ransomware encryption and propagation.

Darktrace’s Cyber AI Analyst automatically investigated and correlated the encryption activity and related events into a single incident narrative, providing the customer with real-time visibility into the attack while significantly reducing investigation time.

Figure 8: Darktrace’s Cyber AI Analyst’s investigation into the encryption activity. AI Analyst incident detailing example encryption activity in real time. Related events are automatically correlated and summarized into a clear narrative, reducing investigation time.

Defender action recommendations

What Could Have Stopped the Attack Earlier?

Although the attack ultimately resulted in ransomware deployment, there were multiple opportunities to detect, contain, and disrupt the intrusion before encryption occurred. The following actions could have significantly reduced the overall impact:

Detect and investigate indicators of reconnaissance and lateral movement

  • Unusual scanning
  • Active Directory replication anomalies consistent with DCSync activity
  • Anomalous use of native tools and processes indicative of LOTL attacks
  • Unusual use of common reconnaissance tools such as Nmap and NetScan

Contain compromised credentials and affected devices

  • Disable and reset compromised VPN credentials
  • Isolate devices performing anomalous scanning and lateral movement activity

Block suspicious external communications and data exfiltration

  • Use anomaly-based detection to detect and block repeated outbound connections to rare external infrastructure
  • Prevent data exfiltration to unauthorized cloud storage services such as Wasabi

Conclusion

The incident highlights the importance of anomaly-based detection, particularly against attacks that primarily use native or legitimate tools to evade traditional security measures. Darktrace identified suspicious activity from the first day of the compromise, providing multiple opportunities to disrupt the intrusion before it progressed to lateral movement and data exfiltration.

In this instance, detection was not the limiting factor; response time was. Prompt investigation and containment of devices exhibiting anomalous behavior could have prevented lateral movement, data exfiltration, and ultimately ransomware deployment.

As adversaries increasingly prioritize stealth over custom malware, relying instead on legitimate tools, valid credentials, and trusted infrastructure, traditional signature-based detection becomes less effective. Identifying subtle behavioral deviations early remains critical to disrupting attacks before they escalate into full-scale ransomware incidents.

Credit to Alexandra Evzona (Cyber Analyst), Priya Thapa (Senior Cyber Analyst)
Edited by Ryan Traill (Content Manager)

Appendices

References

[1] https://industrialcyber.co/ransomware/check-point-reports-ransomware-attacks-jump-48-year-over-year-despite-decline-in-overall-cyberattack-activity/

[2] https://www.europol.europa.eu/media-press/newsroom/news/law-enforcement-disrupt-worlds-biggest-ransomware-operation

[3] https://fieldeffect.com/blog/field-effect-mitigates-not-so-simplehelp-exploits-enabling-deployment-of-backdoors

[4] https://www.darktrace.com/blog/from-vps-to-phishing-how-darktrace-uncovered-saas-hijacks-through-virtual-infrastructure-abuse

Indicators of Compromise (IoCs)

IP Addresses

  • 137[.]220[.]59[.]55 – Potential C2 infrastructure (Vultr AS20473)
  • 38[.]27[.]106[.]123 – Wasabi cloud storage endpoint associated with potential data exfiltration
  • 38[.]27[.]106[.]128 – Wasabi cloud storage endpoint associated with potential data exfiltration
  • 38[.]27[.]106[.]117 – Wasabi cloud storage endpoint associated with potential data exfiltration

Domains

  • safedata[.]s3[.]wasabisys[.]com – Potential data exfiltration endpoint
  • *.wasabisys[.]com – Associated Wasabi cloud storage infrastructure

JA3 Fingerprint

  • d6828e30ab66774a91a96ae93be4ae4c – Associated with the Sliver adversary emulation framework

Files

  • Delete[.]me – File observed during reconnaissance activity, commonly associated with NetScan

Darktrace Model Coverage

·       Anomalous Connection / Active Remote Desktop Tunnel

·       Anomalous Connection / Anomalous Remote Registry Service Control

·       Anomalous Connection / Multiple Failed Windows UDP

·       Anomalous Connection / New or Uncommon Service Control

·       Anomalous Connection / New or Uncommon Service Enumeration

·       Anomalous Connection / New User Agent to IP Without Hostname

·       Anomalous Connection / SMB Enumeration

·       Anomalous Connection / Suspicious Activity On High Risk Device

·       Anomalous Connection / Suspicious Read Write Ratio

·       Anomalous Connection / Sustained MIME Type Conversion

·       Anomalous Connection / Uncommon 1 GiB Outbound

·       Anomalous Connection / Unusual Admin RDP Session

·       Anomalous Connection / Unusual Admin SMB Session

·       Anomalous Connection / Unusual SMB Version 1 Connectivity

·       Anomalous File / EXE from Rare External Location

·       Anomalous File / Internal / Additional Extension Appended to SMB File

·       Anomalous Server Activity / Outgoing from Server

·       Compromise / Beaconing Activity To External Rare

·       Compromise / Ransomware / Possible Ransom Note Read

·       Compromise / Ransomware / Ransom or Offensive Words Written to SMB

·       Compromise / Ransomware / Suspicious SMB Activity

·       Device / Anonymous NTLM Logins

·       Device / Attack and Recon Tools

·       Device / Initial Attack Chain Activity

·       Device / Large Number of Model Alerts

·       Device / Long Agent Connection to New Endpoint

·       Device / Multiple Lateral Movement Model Alerts

·       Device / Network Scan

·       Device / New or Uncommon SMB Named Pipe

·       Device / New or Unusual Remote Command Execution

·       Device / New User Agent

·       Device / Possible RPC Lateral Movement

·       Device / Possible SMB/NTLM Brute Force

·       Device / RDP Scan

·       Device / SMB Lateral Movement

·       Device / SMB Session Brute Force (Non-Admin)

·       Device / SMB Version 1 Access Failures

·       Device / Suspicious File Delete Activity

·       Device / Suspicious Network Scan Activity

·       Device / Unusual SMB Error Detected

·       Device / Unusual SMB To Critical Resource

·       Device / Unusual Winreg Operation

·       Unusual Activity / Multiple Failed Internal Connections

·       Unusual Activity / Possible RPC Recon Activity

·       Unusual Activity / Sustained Anomalous SMB Activity

·       Unusual Activity / Unusual External Data to New Endpoint

·       Unusual Activity / Unusual External Data Transfer

·       Unusual Activity / Unusual File Storage Data Transfer

·       Unusual Activity / Unusual Large Internal Transfer

·       User / New Admin Credentials on Client

·       User / NTLM Login from Unauthenticated Device

Autonomous Response Model Alerts

·       Antigena / Network / External Threat / Antigena Ransomware Block

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

·       Antigena / Network / Insider Threat / Antigena Active Threat SMB Write Block

·       Antigena / Network / Insider Threat / Antigena Internal Anomalous File Activity

·       Antigena / Network / Insider Threat / Antigena Large Data Volume Outbound Block

·       Antigena / Network / Insider Threat / Antigena Network Scan Block

·       Antigena / Network / Insider Threat / Antigena Unusual Privileged User Activities Block

·       Antigena / Network / Manual / Quarantine Device

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

·       Antigena / Network / Significant Anomaly / Antigena Controlled and Model Alert

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

·       Antigena / Network / Significant Anomaly / Antigena Enhanced Monitoring from Server Block 100

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

MITRE ATT&CK MAPPING

Command and control - Protocol Tunneling - T1572

Command and control – Web Protocols – T1071.001

Credential Access- Exploitation for Credential Access- T1212

Credential Access- Password Guessing- T1110

Discovery- File and Directory Discovery- T1083

Discovery- Network Service Discovery- T1046

Discovery- Network Share Discovery- T1135

Discovery- Remote System Discovery- T1018

Exfiltration- Exfiltration Over C2 Channel- T1041

Exfiltration- Exfiltration to Cloud Storage- T1567

Impact- Data Encrypted for Impact- T1486

Impact- Data Encrypted for Impact- T1486

Impact- Service Stop - T1489

Initial Access- Public-Facing Application- T1190

Lateral Movement- Exploitation of Remote Services- T1210

Lateral Movement- Remote Desktop Protocol- T1021

Lateral Movement- SMB/Windows Admin Shares- T1021

Lateral Movement- Taint Shared Content- T1080

Persistence- Modify Registry- T1112

Privilege Escalation- Exploitation for Privilege Escalation- T0890

Privilege Escalation- Valid Accounts- T1078

Reconnaissance- Scanning IP Blocks- T1595

Reconnaissance- Vulnerability Scanning- T1595

Resource Development- Malware- T1588

Stealth- File Deletion- T1070

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