Blog
/
/
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
Default blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog imageDefault 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

/

/

September 23, 2025

It’s Time to Rethink Cloud Investigations

Default blog imageDefault blog image

Cloud Breaches Are Surging

Cloud adoption has revolutionized how businesses operate, offering speed, scalability, and flexibility. But for security teams, this transformation has introduced a new set of challenges, especially when it comes to incident response (IR) and forensic investigations.

Cloud-related breaches are skyrocketing – 82% of breaches now involve cloud-stored data (IBM Cost of a Data Breach, 2023). Yet incidents often go unnoticed for days: according to a 2025 report by Cybersecurity Insiders, of the 65% of organizations experienced a cloud-related incident in the past year, only 9% detected it within the first hour, and 62% took more than 24 hours to remediate it (Cybersecurity Insiders, Cloud Security Report 2025).

Despite the shift to cloud, many investigation practices remain rooted in legacy on-prem approaches. According to a recent report, 65% of organizations spend approximately 3-5 days longer when investigating an incident in the cloud vs. on premises.

Cloud investigations must evolve, or risk falling behind attackers who are already exploiting the cloud’s speed and complexity.

4 Reasons Cloud Investigations Are Broken

The cloud’s dynamic nature – with its ephemeral workloads and distributed architecture – has outpaced traditional incident response methods. What worked in static, on-prem environments simply doesn’t translate.

Here’s why:

  1. Ephemeral workloads
    Containers and serverless functions can spin up and vanish in minutes. Attackers know this as well – they’re exploiting short-lived assets for “hit-and-run” attacks, leaving almost no forensic footprint. If you’re relying on scheduled scans or manual evidence collection, you’re already too late.
  2. Fragmented tooling
    Each cloud provider has its own logs, APIs, and investigation workflows. In addition, not all logs are enabled by default, cloud providers typically limit the scope of their logs (both in terms of what data they collect and how long they retain it), and some logs are only available through undocumented APIs. This creates siloed views of attacker activity, making it difficult to piece together a coherent timeline. Now layer in SaaS apps, Kubernetes clusters, and shadow IT — suddenly you’re stitching together 20+ tools just to find out what happened. Analysts call it the ‘swivel-chair Olympics,’ and it’s burning hours they don’t have.
  3. SOC overload
    Analysts spend the bulk of their time manually gathering evidence and correlating logs rather than responding to threats. This slows down investigations and increases burnout. SOC teams are drowning in noise; they receive thousands of alerts a day, the majority of which never get touched. False positives eat hundreds of hours a month, and consequently burnout is rife.  
  4. Cost of delay
    The longer an investigation takes, the higher its cost. Breaches contained in under 200 days save an average of over $1M compared to those that linger (IBM Cost of a Data Breach 2025).

These challenges create a dangerous gap for threat actors to exploit. By the time evidence is collected, attackers may have already accessed or exfiltrated data, or entrenched themselves deeper into your environment.

What’s Needed: A New Approach to Cloud Investigations

It’s time to ditch the manual, reactive grind and embrace investigations that are automated, proactive, and built for the world you actually defend. Here’s what the next generation of cloud forensics must deliver:

  • Automated evidence acquisition
    Capture forensic-level data the moment a threat is detected and before assets disappear.
  • Unified multi-cloud visibility
    Stitch together logs, timelines, and context across AWS, Azure, GCP, and hybrid environments into a single unified view of the investigation.
  • Accelerated investigation workflows
    Reduce time-to-insight from hours or days to minutes with automated analysis of forensic data, enabling faster containment and recovery.
  • Empowered SOC teams
    Fully contextualised data and collaboration workflows between teams in the SOC ensure seamless handover, freeing up analysts from manual collection tasks so they can focus on what matters: analysis and response.

Attackers are already leveraging the cloud’s agility. Defenders must do the same — adopting solutions that match the speed and scale of modern infrastructure.

Cloud Changed Everything. It’s Time to Change Investigations.  

The cloud fundamentally reshaped how businesses operate. It’s time for security teams to rethink how they investigate threats.

Forensics can no longer be slow, manual, and reactive. It must be instant, automated, and cloud-first — designed to meet the demands of ephemeral infrastructure and multi-cloud complexity.

The future of incident response isn’t just faster. It’s smarter, more scalable, and built for the environments we defend today, not those of ten years ago.  

On October 9th, Darktrace is revealing the next big thing in cloud security. Don’t miss it – sign up for the webinar.

darktrace live event launch
Continue reading
About the author
Kellie Regan
Director, Product Marketing - Cloud Security

Blog

/

/

September 22, 2025

Understanding the Canadian Critical Cyber Systems Protection Act

Default blog imageDefault blog image

Introduction: The Canadian Critical Cyber Systems Protection Act

On 18 June 2025, the Canadian federal Government introduced Bill C-8 which, if adopted following completion of the legislative process, will enact the Critical Cyber Systems Protection Act (CCSPA) and give Canada its first federal, cross-sector and legally binding cybersecurity regime for designated critical infrastructure providers. As of August 2025, the Bill has completed first reading and stands at second reading in the Canadian House of Commons.

Political context

The measure revives most of the stalled 2022 Bill C-26 “An Act Respecting Cyber Security” which “died on Paper” when Parliament was prorogued in January 2025, in the wake of former Prime Minister Justin Trudeau’s resignation.

The new government, led by Mark Carney since March 2025, has re-tabled the package with the same two-part structure: (1) amendments to the Telecommunications Act that enable security directions to telecoms; and (2) a new CCSPA setting out mandatory cybersecurity duties for designated operators. This blog focuses on the latter.

If enacted, Canada will join fellow Five Eyes partners such as the United Kingdom and Australia, which already impose statutory cyber-security duties on operators of critical national infrastructure.

The case for new cybersecurity legislation in Canada

The Canadian cyber threat landscape has expanded. The country's national cyber authority, the Canadian Centre for Cybersecurity (Cyber Centre), recently assessed that the number of cyber incidents has “sharply increased” in the last two years, as has the severity of those incidents, with essential services providers among the targets. Likewise, in its 2025-2026 National Cyber Threat Assessment, the Cyber Centre warned that AI technologies are “amplifying cyberspace threats” by lowering barriers to entry, improving the speed and sophistication of social-engineering attacks and enabling more precise operations.

This context mirrors what we are seeing globally: adversaries, including state actors, are taking advantage of the availability and sophistication of AI tools, which they have leverage to amplify the effectiveness of their operations. In this increasingly complex landscape, regulation must keep pace and evolve in step with the risk.

What the Canadian Critical Cyber Systems Protection Act aims to achieve

  • If enacted, the CCSPA will apply to operators in federally regulated critical infrastructure sectors which are vital to national security and public safety, as further defined in “Scope” below (the “Regulated Entities”), to adopt and comply with a minimum standard of cybersecurity duties (further described below)  which align with those its Five Eyes counterparts are already adhering to.

Who does the CCSPA apply to

The CCSPA would apply to designated operators that deliver services or systems within federal jurisdiction in the following priority areas:

  • telecommunications services
  • interprovincial or international pipeline and power line systems, nuclear energy systems, transportation systems
  • banking and clearing  
  • settlement systems

The CCSPA would also grant the Governor in Council (Federal Cabinet) with powers to add or remove entities in scope via regulation.

Scope of the CCSPA

The CCSPA introduces two key instruments:

First, it strengthens cyber threat information sharing between responsible ministers, sector regulators, and the Communications Security Establishment (through the Cyber Centre).

Second, it empowers the Governor in Council (GIC) to issue Cyber Security Directions (CSDs) - binding orders requiring a designated operator to implement specified measures to protect a critical cyber system within defined timeframes.

CSDs may be tailored to an individual operator or applied to a class of operators and can address technology, process, or supplier risks. To safeguard security and commercial confidentiality, the CCSPA restricts disclosure of the existence or content of a CSD except as necessary to carry it out.

Locating decision-making with the GIC ensures that CSDs are made with a cross-government view that weighs national security, economic priorities and international agreement.

New obligations for designated providers

The CCSPA would impose key cybersecurity compliance and obligations on designated providers. As it stands, this includes:

  1. Establishing and maintaining cybersecurity programs: these will need to be comprehensive, proportionate and developed proactively. Once implemented, they will need to be continuously reviewed
  2. Mitigating supply chain risks: Regulated Entities will be required to assess their third-party products and services by conducting a supply chain analysis, and take active steps to mitigate any identified risks
  3. Reporting incidents:  Regulated Entities will need to be more transparent with their reporting, by making the Communications Security Establishment (CSE) aware of any incident which has, or could potentially have, an impact on a critical system. The reports must be made within specific timelines, but in any event within no more than 72 hours;
  4. Compliance with cybersecurity directions:  the government will, under the CCSPA, have the authority to issue cybersecurity directives in an effort to remain responsive to emerging threats, which Regulated Entities will be required to follow once issued
  5. Record keeping: this shouldn’t be a surprise to many of those Regulated Entities which fall in scope, which are already likely to be subject to record keeping requirements. Regulated Entities should expect to be maintaining records and conducting audits of their systems and processes against the requirements of the CCSPA

It should be noted, however, that this may be subject to change, so Regulated Entities should keep an eye on the progress of the Bill as it makes its way through parliament.

Enforcement of the Act would be carried out by sector-specific regulators identified in the Act such as the Office of the Superintendent of Financial Institutions, Minister of Transport, Canada Energy Regulator, Canadian Nuclear Safety Commission and the Ministry of Industry.

What are the penalties for CCSPA non-compliance?

When assessing the penalties associated with non-compliance with the requirements of the CCSPA, it is clear that such non-compliance will be taken seriously, and the severity of the penalties follows the trend of those applied by the European Union to key pieces of EU legislation. The “administrative monetary penalties” (AMPs) set by regulation could see fines being applied of up to C$1 million for individuals and up to C$15 million for organizations.

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