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.  

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.

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 2.78 billion 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.

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]

With 90million investigations in 2024 alone, Darktrace Cyber AI Analyst ^TM is transforming security operations with AI and has added up to 30 Full Time Security Analysts to almost 10,000 security teams.

In today’s high-stakes threat landscape, security teams are overwhelmed — stretched thin by burnout, alert fatigue, and a constant barrage of fast-moving attacks. As traditional tools can’t keep up, many are turning to AI to solve these challenges. But not all AI is created equal, and no single type of AI can perform all the functions necessary to effectively streamline security operations, safeguard your organization and rapidly respond to threats.

Thus, a multi-layered AI approach is critical to enhance threat detection, investigation, and response and augment security teams. By leveraging multiple AI methods, such as machine learning, deep learning, and natural language processing, security systems become more adaptive and resilient, capable of identifying and mitigating complex cyber threats in real time. This comprehensive approach ensures that no single AI method's limitations compromise the overall security posture, providing a robust defense against evolving threats.

As leaders in AI in cybersecurity, Darktrace has been utilizing a multi-layered AI approach for years, strategically combining and layering a range of AI techniques to provide better security outcomes. One key component of this is our Cyber AI Analyst – a sophisticated agentic AI system that avoids the pitfalls of generative AI. This approach ensures expeditious and scalable investigation and analysis, accurate threat detection and rapid automated response, empowering security teams to stay ahead of today's sophisticated cyber threats.

In this blog we will explore:

• What agentic AI is and why security teams are adopting it to deliver a set of critical functions needed in cybersecurity
• How Darktrace’s Cyber AI Analyst^TM is a sophisticated agentic AI system that uses a multi-layered AI approach to achieve better security outcomes and enhance SOC analysts
• Introduce two new innovative machine learning models that further augment Cyber AI Analyst’s investigation and evaluation capabilities

The rise of agentic AI

To combat the overwhelming volume of alerts, the shortage of security professionals, and burnout, security teams need AI that can perform complex tasks without human intervention, also known as agentic AI. The ability of these systems to act autonomously can significantly improve efficiency and effectiveness. However, many attempts to implement agentic AI rely on generative AI, which has notable drawbacks.

Broadly speaking, agentic AI refers to artificial intelligence systems that act autonomously as "agents," capable of carrying out complex tasks, making decisions, and interacting with tools or external systems with no or limited human intervention. Unlike traditional AI models that perform predefined tasks, it uses advanced techniques to mimic human decision-making processes, dynamically adapting to new challenges and responding to varied inputs. In a narrower definition, agentic AI often uses generative large language models (LLMs) as its core, using this to plan tasks and interactions with other systems, iteratively feeding its output into its input to accomplish more tasks than are traditionally possible with a single prompt. When described in terms of technology rather than functionality, agentic AI would be deemed as AI using this kind of generative system.

In cybersecurity, agentic AI systems can be used to autonomously monitor traffic, identify unusual patterns or anomalies indicating potential threats, and take action to respond to these possible attacks. For example, they can handle incident response tasks such as isolating affected systems or patching vulnerabilities, and triaging alerts. This reduces the reliance on human analysts for routine tasks, allowing them to focus on high-priority incidents and strategic initiatives, thereby increasing the overall efficiency and effectiveness of the SOC.

Despite their potential, agentic AI systems with a generative AI core have notable limitations. Whether based on widely used foundation models or fully custom proprietary implementations, generative AI often struggles with poor reasoning and can produce incorrect conclusions. These models are prone to "hallucinations," where they generate false information, which can be magnified through iterative processes. Additionally, generative AI systems are particularly susceptible to inheriting biases from training data, leading to incorrect outcomes, and are vulnerable to adversarial attacks, such as prompt injection that manipulates the AI's decision-making process.

Thus, choosing the right agentic AI system is crucial for security teams to ensure accurate threat detection, streamline investigations, and minimize false positives. It's essential to look beyond generative AI-based systems, which can lead to false positives and missed threats, and adopt AI that integrates multiple techniques. By considering AI systems that leverage a variety of advanced methods, organizations can build a more robust and comprehensive security strategy.  

Industry’s most experienced agentic AI analyst

First introduced in 2019, Darktrace Cyber AI Analyst^TM emerged as a groundbreaking, patented solution in the cybersecurity landscape. As the most experienced AI Analyst deployed to almost 10,000 customers worldwide, Cyber AI Analyst is a sophisticated example of agentic AI, aligning closely with our broad definition. Unlike generative AI-based systems, it uses a multi-layered AI approach - strategically combining and layering various AI techniques, both in parallel and sequentially – to autonomously investigate and triage alerts with speed and precision that outpaces human teams. By utilizing a diverse set of AI methods, including unsupervised machine learning, models trained on expert cyber analysts, and custom security-specific large language models, Cyber AI Analyst mirrors human investigative processes by questioning data, testing hypotheses, and reaching conclusions at machine speed and scale. It integrates data from various sources – including network, cloud, email, OT and even third-party alerts – to identify threats and execute appropriate responses without human input, ensuring accurate and reliable decision-making.

With its ability to learn and adapt using Darktrace's unique understanding of an organization’s environment, Cyber AI Analyst highlights anomalies and passes only the most relevant activity to human users. Every investigation is thoroughly explained with natural language summaries, providing transparent and interpretable AI insights. Unlike generative AI-based agentic systems, Cyber AI Analyst's outputs are based on a comprehensive understanding of the underlying data, avoiding inaccuracies and "hallucinations," thereby dramatically reducing risk of false positives.

90 million investigations. Zero burnout.

Building on six years of innovation since launch, Darktrace's Cyber AI Analyst continues to revolutionize security operations by automating time-consuming tasks and enabling teams to focus on strategic initiatives. In 2024 alone, the sophisticated AI system autonomously conducted 90 million investigations, its analysis and correlation during these investigations resulted in escalating just 3 million incidents for human validation and resulting in fewer than 500,000 incidents deemed critical to the security of the organization. This completely changed the security operations process, providing customers with an ability to investigate every relevant alert as an unprecedented alternative to detection engineering that avoids massive quantities of risk from the traditional approach.  Cyber AI Analyst performed the equivalent of 42 million hours of human investigation for relevant security alerts.

The benefits of Cyber AI Analyst will transform security operations as we know it today:

• Autonomously investigates thousands of alerts, distilling them into a few critical incidents — saving security teams thousands of hours and removing risk from current “triage few” processes. [See how the State of Oklahoma gained 2,561 hours of investigation time and eliminated 3,142 alerts in 3 months]
• It decreases critical incident discoverability from hours to minutes, enabling security teams to respond faster to potential threats that will severely impact their organization. Learn how South Coast Water District went from hours to minutes in incident discovery.
• It reduces false positives by 90%, giving security teams confidence in its accuracy and output.
• Delivers the output of up to 30 full-time analysts – without the cost, burnout, or ramp-up time, while elevating existing human security analysts to validation and response

Cyber AI Analyst allows security teams to allocate their resources more effectively, focusing on genuine threats rather than sifting through noise. This not only enhances productivity but also ensures that critical alerts are addressed promptly, minimizing potential damage and improving overall cyber resilience.

Always innovating - Next-generation AI models for cybersecurity

As empowering defenders with AI has never been more critical, Darktrace remains committed to driving innovation that helps our customers proactively reduce risk, strengthen their security posture, and uplift their teams. To further enhance security teams, Darktrace is introducing two next-generation AI models for cybersecurity within Cyber AI Analyst, including:

• Darktrace Incident Graph Evaluation for Security Threats (DIGEST): Using graph neural networks, this model analyzes how attacks progress to predict which threats are likely to escalate — giving your team earlier warnings and sharper prioritization.  This means earlier warnings, better prioritization, and fewer surprises during active threats.
• Darktrace Embedding Model for Investigation of Security Threats - Version 2 (DEMIST-2): This new language model is purpose-built for cybersecurity. With deep contextual understanding, it automates critical human-like analysis— like assessing hostnames, file sensitivity, and tracking users across environments. Unlike large general-purpose models, it delivers superior performance with a smaller footprint. Working across all our deployment types, including on-prem and cloud, it can run without internet access, keeping inference local.

Unlike the foundational LLMs that power many generative and agentic systems, these models are purpose-built for cybersecurity, supported by insights of over 200 security analysts and is capable of mimicking how an analyst thinks, to bring AI-based precision and depth of analysis into the SOC. By understanding how attacks evolve and predicting which threats are most likely to escalate, these machine learning models enable Cyber AI Analyst^TM to provide earlier detection, sharper prioritization, and faster, more confident decision-making.

Conclusion

Darktrace Cyber AI Analyst^TM redefines security operations with proven agentic AI — delivering autonomous investigations and faster response times, while significantly reducing false positives. With powerful new models like DIGEST and DEMIST-2, it empowers security teams to prioritize what matters, cut through noise, and stay ahead of evolving threats — all without additional headcount. As cyber risk grows, Cyber AI Analyst stands out as a force multiplier, driving efficiency, resilience, and confidence in every SOC.

[related-resource]

Additional resources

• Learn more about Darktrace’s unique approach to AI by downloading “The AI Arsenal: Understanding the Tools Shaping Cybersecurity.”
• Connect with Darktrace at RSAC 2025 in San Francisco April 28– May 1^st at Booth S-2227 or book a one-on-one meeting with Darktrace experts on-site here.
• Discover the principles guiding Darktrace’s development of responsible AI by downloading “Towards Responsible AI in Cybersecurity.”