Blog
/
/
August 4, 2020

How to Prevent Spear Phishing Attacks Post Twitter Hack

Twitter confirmed spear phishing as the cause of last month's attack. Learn about the limits of current defenses against spear phishing and how AI can stop it.
Inside the SOC
Darktrace cyber analysts are world-class experts in threat intelligence, threat hunting and incident response, and provide 24/7 SOC support to thousands of Darktrace customers around the globe. Inside the SOC is exclusively authored by these experts, providing analysis of cyber incidents and threat trends, based on real-world experience in the field.
Written by
Dan Fein
VP, Product
Default blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog image
04
Aug 2020

Twitter has now confirmed that it was a “phone spear phishing attack” targeting a small number of their employees that allowed hackers to access 130 high-profile user accounts and fool thousands of people into giving away money via bitcoin.

Spear phishing involves targeted texts or emails aimed at individuals in an attempt to ‘hook’ them into opening an attachment or malicious link. This attack highlights the limitations in the security controls adopted by even some of the largest and most tech-savvy organizations out there, who continue to fall victim to this well-known attack technique.

The incident has been described by Twitter as a “coordinated social engineering attack” that “successfully targeted employees with access to internal systems and tools.”

Though the specific nature of the attack remains unclear, it likely followed a similar pattern to the series of threat finds detailed elsewhere on the Darktrace Blog: impersonating trusted colleagues or platforms, such as WeTransfer, Microsoft Teams or even Twitter itself, with an urgent message coaxing an employee into clicking on a disguised URL and inputting their credentials on a fake login page.

When an employee inputs their credentials, that data is recorded and beaconed back to the attacker, who will then use these login details to access internal systems — which, in this case, allowed them to subsequently take control of celebrities’ Twitter accounts and send out the damaging Tweets that left thousands out of pocket.

Training the workforce is not enough

Twitter says in a statement that this incident has forced them to “accelerate several of [their] pre-existing security workstreams.” But the suggestion that they will continue to organize “ongoing company-wide phishing exercises throughout the year” indicates an over-reliance on the ability of humans to identify these malicious email attacks that are getting more and more advanced, and harder to distinguish from genuine communication.

Cyber-criminals are now using AI to create fake profiles, personalize messages and replicate communication patterns, at a speed and scale that no human ever could. In this threat landscape, there can no longer be a reliance solely on educating the workforce, as the difference between a malicious email and legitimate communication becomes almost imperceptible. This has led to an acceptance that we must rely on technology to help us catch the subtle signs of attack, when humans alone fail to do so.

The legacy approach: no playbook for new attacks

The majority of communications security systems are not where they need to be, and this is particularly true for the email realm. Most tools in use today rely on static blacklists of rules and signatures that analyze emails in isolation, against known ‘bads’. Methods like looking for IP addresses or file hashes associated with phishing have had limited success in stopping attackers, who have devised simple techniques to bypass them.

As we have explored previously, attackers are constantly changing their approach, purchasing new domains en masse, experimenting with novel strains of malware, and manipulating headers to get around common validation checks. It is due to these developments that Secure Email Gateways (SEGs) become antiquated almost the moment they are updated.

The mean lifetime of an attack has reduced from 2.1 days in 2018 to 0.5 days in 2020. As soon as an SEG identifies a domain or a file hash as malicious, cyber-criminals change their attack infrastructure and launch a new wave of fresh attacks. Their fundamental means of operation renders legacy security tools incapable of evolving with the threat landscape, and it is for this reason that over 94% of cyber-attacks today start with an email.

How Cyber AI catches the threats others miss

However, one area where email security has seen great progress even in the last two years is the application of AI to spot the subtle features of advanced email attacks, even those that leverage novel malware. This approach allows security tools to move away from the binary decision-making that comes with asking “Is this email ‘bad’?” and moving to the far more useful question of “does this belong?”

This form of what we’re calling ‘layered AI’ combines supervised and unsupervised machine learning, enabling it to spot the subtle deviations from learned ‘patterns of life’ that are indicative of a cyber-threat.

Supervised machine learning models can be trained on millions of emails to find subtle patterns undetectable by humans and detect new variations of known threat types. These models are able to find the real-world intentions behind an email: by training on millions of spear phishing emails, for example, a system can find patterns associated with this type of email attack and accurately classify a future email as spear phishing.

In addition, unsupervised machine learning models can be trained on all available email data for an organization to find unknown variations of unknown threat types — that is, the ‘unknown unknowns,’ the combinations never before seen. Ultimately this is what enables a system to ask that critical question “does this belong?” and spot genuine anomalies that fall outside of the norm.

Layering both of these applications of AI allows us to make determinations such as: ‘this is a phishing email and it doesn’t belong’, dramatically improving the system’s accuracy and allowing it to interrupt only the malicious emails – since there could be phishy-looking emails that are legitimate! It also enables us to act in proportion to the threat identified: locking links and attachments in some cases, or holding back emails entirely in others.

This form of ‘layered AI’ requires an advanced understanding of mathematics and machine learning that takes years of research and development. With that experience, Cyber AI has proven itself capable of catching the full range of advanced attacks targeting the inbox, from spear phishing and impersonation attempts, to account takeovers and supply chain attacks. Once implemented, it takes only a week before any new organization can derive value, and thousands of customers now rely on Cyber AI to protect both their email realm and wider network.

Plenty more phish in the sea

This will not be the last time this year that a cyber-attack caused by spear phishing makes the headlines. Just this week, it was revealed that Russian-backed cyber-criminals stole sensitive documents on US-UK trade talks after successful spear phishing, and the technique may well have played a part in ongoing vaccine research espionage that surfaced in July.

With the US presidential race heating up, it was recently revealed that fewer than 3 out of 10 election administrators have basic controls to prevent phishing. This attack method may come to not only damage organizations and their reputation, but also to undermine the trust that serves as the bedrock of democracy. Now is the time to start recognizing the very real threat that email attackers represent, and to prepare our defenses accordingly.

Learn more about AI email security

Inside the SOC
Darktrace cyber analysts are world-class experts in threat intelligence, threat hunting and incident response, and provide 24/7 SOC support to thousands of Darktrace customers around the globe. Inside the SOC is exclusively authored by these experts, providing analysis of cyber incidents and threat trends, based on real-world experience in the field.
Written by
Dan Fein
VP, Product

More in this series

No items found.

Blog

/

/

April 21, 2025

Why Asset Visibility and Signature-Based Threat Detection Fall Short in ICS Security

operational technology operators looking at equipment Default blog imageDefault blog image

In the realm of Industrial Control System (ICS) security, two concepts often dominate discussions:

  1. Asset visibility
  2. Signature-based threat detection

While these are undoubtedly important components of a cybersecurity strategy, many organizations focus on them as the primary means to enhance ICS security. However, this is more of a short-term approach and these organizations often realize too late that these efforts do not translate into actually securing their environment.

To truly secure your environment, organizations should focus their efforts on anomaly detection across core network segments. This shift enables enhanced threat detection, while also providing a more meaningful and dynamic view of asset communication.

By prioritizing anomaly detection, organizations can build a more resilient security posture, detecting and mitigating threats before they escalate into serious incidents.

The shortcomings of asset visibility and signature-based threat detection

Asset visibility is frequently touted as the foundation of ICS security. The idea is that you cannot protect what you cannot see.

However, organizations that invest heavily in asset discovery tools often end up with extensive inventories of connected devices but little actionable insight into their security posture or risk level, let alone any indication as to whether these assets have been compromised.

Simply knowing what assets exist does not equate to securing them.

Worse, asset discovery is often a time-consuming static process. By the time practitioners complete their inventory, not only is there likely to have been changes to their assets, but the threat landscape may have already evolved, introducing new vulnerabilities and attack vectors  that were not previously accounted for.

Signature-based detection is reactive, not proactive

Traditional signature-based threat detection relies on known attack patterns and predefined signatures to identify malicious activity. This approach is fundamentally reactive because it can only detect threats that have already been identified elsewhere.

In an ICS environment where cyber-attacks on OT systems have become more frequent, sophisticated, and destructive, signature-based detection provides a false sense of security while failing to detect sophisticated, previously unseen threats:

Additionally, adversaries often dwell within OT networks for extended periods, studying their specific conditions to identify the most effective way to cause disruption. This means that the likelihood of any attack within OT network looking the same as a previous attack is unlikely.

Implementation effort vs. actual security gains

Many organizations spend considerable time and resources implementing asset visibility solutions and signature-based detection systems only to be required to constantly tune and adjust the sensitivity of the solution.

Despite these efforts, these tools often fail to deliver the level of protection expected, leaving gaps in detection, an overwhelming amount of asset data, and a constant stream of false positives and false negatives from signature-based systems.

A more effective approach: Anomaly detection at core network segments

While it's important to understand the type of device involved during alert triage, organizations should shift their focus from static asset visibility and threat signatures to anomaly detection across critical network segments. This method provides a superior approach to ICS security for several reasons:

Proactive threat detection

Anomaly detection monitors network behavior in real time and identifies deviations . This means that even novel or previously unseen threats can be detected based on unusual network activity, rather than relying on predefined signatures.

Granular security insights

By analyzing traffic patterns across key network segments, organizations can gain deeper insights into how assets interact. This not only improves threat detection but also organically enhances asset visibility. Instead of simply cataloging devices, organizations gain meaningful visibility into how they behave within the network, understanding their unique pattern of life, and making it easier to detect malicious activity.

Efficiency and scalability

Implementing anomaly detection allows security teams to focus on real threats rather than sifting through massive inventories of assets or managing signature updates. It scales better with evolving threats and provides continuous monitoring without requiring constant manual intervention.

Enhanced threat detection for critical infrastructure

Unlike traditional security approaches that rely on static baselines or threat intelligence that doesn't reflect the unique behaviors of your OT environment, Darktrace / OT uses multiple AI techniques to continuously learn and adapt to your organization’s real-world activity across IT, OT, and IoT.

By building a dynamic understanding of each device’s pattern of life, it detects threats at every stage of the kill chain — from known malware to zero-days and insider attacks — without overwhelming your team with false positives or unnecessary alerts. This ensures scalable protection as your environment evolves, without a significant increase in operational overhead.

[related-resource]

Continue reading
About the author
Jeffrey Macre
Industrial Security Solutions Architect

Blog

/

/

April 16, 2025

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

woman looking at laptop at deskDefault blog imageDefault blog image

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 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.

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]

Continue reading
About the author
Margaret Cunningham, PhD
Director, Security & AI Strategy, Field CISO
Your data. Our AI.
Elevate your network security with Darktrace AI