The State of AI in Cybersecurity: Understanding AI Technologies
Part 4: This blog explores the findings from Darktrace’s State of AI Cybersecurity Report on security professionals' understanding of the different types of AI used in security programs. Get the latest insights into the evolving challenges, growing demand for skilled professionals, and the need for integrated security solutions by downloading the full report.
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
The Darktrace Community
Share
24
Jul 2024
About the State of AI Cybersecurity Report
Darktrace surveyed 1,800 CISOs, security leaders, administrators, and practitioners from industries around the globe. Our research was conducted to understand how the adoption of new AI-powered offensive and defensive cybersecurity technologies are being managed by organizations.
How familiar are security professionals with supervised machine learning
Just 31% of security professionals report that they are “very familiar” with supervised machine learning.
Many participants admitted unfamiliarity with various AI types. Less than one-third felt "very familiar" with the technologies surveyed: only 31% with supervised machine learning and 28% with natural language processing (NLP).
Most participants were "somewhat" familiar, ranging from 46% for supervised machine learning to 36% for generative adversarial networks (GANs). Executives and those in larger organizations reported the highest familiarity.
Combining "very" and "somewhat" familiar responses, 77% had familiarity with supervised machine learning, 74% generative AI, and 73% NLP. With generative AI getting so much media attention, and NLP being the broader area of AI that encompasses generative AI, these results may indicate that stakeholders are understanding the topic on the basis of buzz, not hands-on work with the technologies.
If defenders hope to get ahead of attackers, they will need to go beyond supervised learning algorithms trained on known attack patterns and generative AI. Instead, they’ll need to adopt a comprehensive toolkit comprised of multiple, varied AI approaches—including unsupervised algorithms that continuously learn from an organization’s specific data rather than relying on big data generalizations.
Different types of AI
Different types of AI have different strengths and use cases in cyber security. It’s important to choose the right technique for what you’re trying to achieve.
Supervised machine learning: Applied more often than any other type of AI in cyber security. Trained on human attack patterns and historical threat intelligence.
Large language models (LLMs): Applies deep learning models trained on extremely large data sets to understand, summarize, and generate new content. Used in generative AI tools.
Natural language processing (NLP): Applies computational techniques to process and understand human language.
Unsupervised machine learning: Continuously learns from raw, unstructured data to identify deviations that represent true anomalies.
What impact will generative AI have on the cybersecurity field?
More than half of security professionals (57%) believe that generative AI will have a bigger impact on their field over the next few years than other types of AI.
Figure 1: Chart from Darktrace's State of AI in Cybersecurity Report
Security stakeholders are highly aware of generative AI and LLMs, viewing them as pivotal to the field's future. Generative AI excels at abstracting information, automating tasks, and facilitating human-computer interaction. However, LLMs can "hallucinate" due to training data errors and are vulnerable to prompt injection attacks. Despite improvements in securing LLMs, the best cyber defenses use a mix of AI types for enhanced accuracy and capability.
AI education is crucial as industry expectations for generative AI grow. Leaders and practitioners need to understand where and how to use AI while managing risks. As they learn more, there will be a shift from generative AI to broader AI applications.
Do security professionals fully understand the different types of AI in security products?
Only 26% of security professionals report a full understanding of the different types of AI in use within security products.
Confusion is prevalent in today’s marketplace. Our survey found that only 26% of respondents fully understand the AI types in their security stack, while 31% are unsure or confused by vendor claims. Nearly 65% believe generative AI is mainly used in cybersecurity, though it’s only useful for identifying phishing emails. This highlights a gap between user expectations and vendor delivery, with too much focus on generative AI.
Key findings include:
Executives and managers report higher understanding than practitioners.
Larger organizations have better understanding due to greater specialization.
As AI evolves, vendors are rapidly introducing new solutions faster than practitioners can learn to use them. There's a strong need for greater vendor transparency and more education for users to maximize the technology's value.
To help ease confusion around AI technologies in cybersecurity, Darktrace has released the CISO’s Guide to Cyber AI. A comprehensive white paper that categorizes the different applications of AI in cybersecurity. Download the White Paper here.
Do security professionals believe generative AI alone is enough to stop zero-day threats?
No! 86% of survey participants believe generative AI alone is NOT enough to stop zero-day threats
This consensus spans all geographies, organization sizes, and roles, though executives are slightly less likely to agree. Asia-Pacific participants agree more, while U.S. participants agree less.
Despite expecting generative AI to have the most impact, respondents recognize its limited security use cases and its need to work alongside other AI types. This highlights the necessity for vendor transparency and varied AI approaches for effective security across threat prevention, detection, and response.
Stakeholders must understand how AI solutions work to ensure they offer advanced, rather than outdated, threat detection methods. The survey shows awareness that old methods are insufficient.
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.
From VPS to Phishing: How Darktrace Uncovered SaaS Hijacks through Virtual Infrastructure Abuse
What is a VPS and how are they abused?
A Virtual Private Server (VPS) is a virtualized server that provides dedicated resources and control to users on a shared physical device. VPS providers, long used by developers and businesses, are increasingly misused by threat actors to launch stealthy, scalable attacks. While not a novel tactic, VPS abuse is has seen an increase in Software-as-a-Service (SaaS)-targeted campaigns as it enables attackers to bypass geolocation-based defenses by mimicking local traffic, evade IP reputation checks with clean, newly provisioned infrastructure, and blend into legitimate behavior [3].
VPS providers like Hyonix and Host Universal offer rapid setup and minimal open-source intelligence (OSINT) footprint, making detection difficult [1][2]. These services are not only fast to deploy but also affordable, making them attractive to attackers seeking anonymous, low-cost infrastructure for scalable campaigns. Such attacks tend to be targeted and persistent, often timed to coincide with legitimate user activity, a tactic that renders traditional security tools largely ineffective.
Darktrace’s investigation into Hyonix VPS abuse
In May 2025, Darktrace’s Threat Research team investigated a series of incidents across its customer base involving VPS-associated infrastructure. The investigation began with a fleet-wide review of alerts linked to Hyonix (ASN AS931), revealing a noticeable spike in anomalous behavior from this ASN in March 2025. The alerts included brute-force attempts, anomalous logins, and phishing campaign-related inbox rule creation.
Darktrace identified suspicious activity across multiple customer environments around this time, but two networks stood out. In one instance, two internal devices exhibited mirrored patterns of compromise, including logins from rare endpoints, manipulation of inbox rules, and the deletion of emails likely used in phishing attacks. Darktrace traced the activity back to IP addresses associated with Hyonix, suggesting a deliberate use of VPS infrastructure to facilitate the attack.
On the second customer network, the attack was marked by coordinated logins from rare IPs linked to multiple VPS providers, including Hyonix. This was followed by the creation of inbox rules with obfuscated names and attempts to modify account recovery settings, indicating a broader campaign that leveraged shared infrastructure and techniques.
Darktrace’s Autonomous Response capability was not enabled in either customer environment during these attacks. As a result, no automated containment actions were triggered, allowing the attack to escalate without interruption. Had Autonomous Response been active, Darktrace would have automatically blocked connections from the unusual VPS endpoints upon detection, effectively halting the compromise in its early stages.
Case 1
Figure 1: Timeline of activity for Case 1 - Unusual VPS logins and deletion of phishing emails.
Initial Intrusion
On May 19, 2025, Darktrace observed two internal devices on one customer environment initiating logins from rare external IPs associated with VPS providers, namely Hyonix and Host Universal (via Proton VPN). Darktrace recognized that these logins had occurred within minutes of legitimate user activity from distant geolocations, indicating improbable travel and reinforcing the likelihood of session hijacking. This triggered Darktrace / IDENTITY model “Login From Rare Endpoint While User Is Active”, which highlights potential credential misuse when simultaneous logins occur from both familiar and rare sources.
Shortly after these logins, Darktrace observed the threat actor deleting emails referring to invoice documents from the user’s “Sent Items” folder, suggesting an attempt to hide phishing emails that had been sent from the now-compromised account. Though not directly observed, initial access in this case was likely achieved through a similar phishing or account hijacking method.
Figure 2: Darktrace / IDENTITY model "Login From Rare Endpoint While User Is Active", which detects simultaneous logins from both a common and a rare source to highlight potential credential misuse.
Case 2
Figure 3: Timeline of activity for Case 2 – Coordinated inbox rule creation and outbound phishing campaign.
In the second customer environment, Darktrace observed similar login activity originating from Hyonix, as well as other VPS providers like Mevspace and Hivelocity. Multiple users logged in from rare endpoints, with Multi-Factor Authentication (MFA) satisfied via token claims, further indicating session hijacking.
Establishing control and maintaining persistence
Following the initial access, Darktrace observed a series of suspicious SaaS activities, including the creation of new email rules. These rules were given minimal or obfuscated names, a tactic often used by attackers to avoid drawing attention during casual mailbox reviews by the SaaS account owner or automated audits. By keeping rule names vague or generic, attackers reduce the likelihood of detection while quietly redirecting or deleting incoming emails to maintain access and conceal their activity.
One of the newly created inbox rules targeted emails with subject lines referencing a document shared by a VIP at the customer’s organization. These emails would be automatically deleted, suggesting an attempt to conceal malicious mailbox activity from legitimate users.
Mirrored activity across environments
While no direct lateral movement was observed, mirrored activity across multiple user devices suggested a coordinated campaign. Notably, three users had near identical similar inbox rules created, while another user had a different rule related to fake invoices, reinforcing the likelihood of a shared infrastructure and technique set.
Privilege escalation and broader impact
On one account, Darktrace observed “User registered security info” activity was shortly after anomalous logins, indicating attempts to modify account recovery settings. On another, the user reset passwords or updated security information from rare external IPs. In both cases, the attacker’s actions—including creating inbox rules, deleting emails, and maintaining login persistence—suggested an intent to remain undetected while potentially setting the stage for data exfiltration or spam distribution.
On a separate account, outbound spam was observed, featuring generic finance-related subject lines such as 'INV#. EMITTANCE-1'. At the network level, Darktrace / NETWORK detected DNS requests from a device to a suspicious domain, which began prior the observed email compromise. The domain showed signs of domain fluxing, a tactic involving frequent changes in IP resolution, commonly used by threat actors to maintain resilient infrastructure and evade static blocklists. Around the same time, Darktrace detected another device writing a file named 'SplashtopStreamer.exe', associated with the remote access tool Splashtop, to a domain controller. While typically used in IT support scenarios, its presence here may suggest that the attacker leveraged it to establish persistent remote access or facilitate lateral movement within the customer’s network.
Conclusion
This investigation highlights the growing abuse of VPS infrastructure in SaaS compromise campaigns. Threat actors are increasingly leveraging these affordable and anonymous hosting services to hijack accounts, launch phishing attacks, and manipulate mailbox configurations, often bypassing traditional security controls.
Despite the stealthy nature of this campaign, Darktrace detected the malicious activity early in the kill chain through its Self-Learning AI. By continuously learning what is normal for each user and device, Darktrace surfaced subtle anomalies, such as rare login sources, inbox rule manipulation, and concurrent session activity, that likely evade traditional static, rule-based systems.
As attackers continue to exploit trusted infrastructure and mimic legitimate user behavior, organizations should adopt behavioral-based detection and response strategies. Proactively monitoring for indicators such as improbable travel, unusual login sources, and mailbox rule changes, and responding swiftly with autonomous actions, is critical to staying ahead of evolving threats.
Credit to Rajendra Rushanth (Cyber Analyst), Jen Beckett (Cyber Analyst) and Ryan Traill (Analyst Content Lead)
• SaaS / Access / Unusual External Source for SaaS Credential Use
• SaaS / Compromise / High Priority Login From Rare Endpoint
• SaaS / Compromise / Login From Rare Endpoint While User Is Active
List of Indicators of Compromise (IoCs)
Format: IoC – Type – Description
• 38.240.42[.]160 – IP – Associated with Hyonix ASN (AS931)
• 103.75.11[.]134 – IP – Associated with Host Universal / Proton VPN
• 162.241.121[.]156 – IP – Rare IP associated with phishing
• 194.49.68[.]244 – IP – Associated with Hyonix ASN
• 193.32.248[.]242 – IP – Used in suspicious login activity / Mullvad VPN
• 50.229.155[.]2 – IP – Rare login IP / AS 7922 ( COMCAST-7922 )
• 104.168.194[.]248 – IP – Rare login IP / AS 54290 ( HOSTWINDS )
• 38.255.57[.]212 – IP – Hyonix IP used during MFA activity
• 103.131.131[.]44 – IP – Hyonix IP used in login and MFA activity
• 178.173.244[.]27 – IP – Hyonix IP
• 91.223.3[.]147 – IP – Mevspace Poland, used in multiple logins
• 2a02:748:4000:18:0:1:170b[:]2524 – IPv6 – Hivelocity VPS, used in multiple logins and MFA activity
• 51.36.233[.]224 – IP – Saudi ASN, used in suspicious login
• 103.211.53[.]84 – IP – Excitel Broadband India, used in security info update
MITRE ATT&CK Mapping
Tactic – Technique – Sub-Technique
• Initial Access – T1566 – Phishing
T1566.001 – Spearphishing Attachment
• Execution – T1078 – Valid Accounts
• Persistence – T1098 – Account Manipulation
T1098.002 – Exchange Email Rules
• Command and Control – T1071 – Application Layer Protocol
T1071.001 – Web Protocols
• Defense Evasion – T1036 – Masquerading
• Defense Evasion – T1562 – Impair Defenses
T1562.001 – Disable or Modify Tools
• Credential Access – T1556 – Modify Authentication Process
T1556.004 – MFA Bypass
• Discovery – T1087 – Account Discovery
• Impact – T1531 – Account Access Removal
The content provided in this blog is published by Darktrace for general informational purposes only and reflects our understanding of cybersecurity topics, trends, incidents, and developments at the time of publication. While we strive to ensure accuracy and relevance, the information is provided “as is” without any representations or warranties, express or implied. Darktrace makes no guarantees regarding the completeness, accuracy, reliability, or timeliness of any information presented and expressly disclaims all warranties.
Nothing in this blog constitutes legal, technical, or professional advice, and readers should consult qualified professionals before acting on any information contained herein. Any references to third-party organizations, technologies, threat actors, or incidents are for informational purposes only and do not imply affiliation, endorsement, or recommendation.
Darktrace, its affiliates, employees, or agents shall not be held liable for any loss, damage, or harm arising from the use of or reliance on the information in this blog.
The cybersecurity landscape evolves rapidly, and blog content may become outdated or superseded. We reserve the right to update, modify, or remove any content without notice.
From Exploit to Escalation: Tracking and Containing a Real-World Fortinet SSL-VPN Attack
Threat actors exploiting Fortinet CVEs
Over the years, Fortinet has issued multiple alerts about a wave of sophisticated attacks targeting vulnerabilities in its SSL-VPN infrastructure. Despite the release of patches to address these vulnerabilities, threat actors have continued to exploit a trio of Common Vulnerabilities and Exposures (CVEs) disclosed between 2022 and 2024 to gain unauthorized access to FortiGate devices.
Which vulnerabilities are exploited?
The vulnerabilities—CVE-2022-42475, CVE-2023-27997, and CVE-2024-21762—affect Fortinet’s SSL-VPN services and have been actively exploited by threat actors to establish initial access into target networks.
The vulnerabilities affect core components of FortiOS, allowing attackers to execute remote code on affected systems.
CVE-2022-42475
Type: Heap-Based Buffer Overflow in FortiOS SSL-VPN
This earlier vulnerability also targets the SSL-VPN interface and has been actively exploited in the wild. It allows attackers to execute arbitrary code remotely by overflowing a buffer in memory, often used to deploy malware or establish persistent backdoors [6].
CVE-2023-27997
Type: Heap-Based Buffer Overflow in FortiOS and FortiProxy
Impact: Remote Code Execution
This flaw exists in the SSL-VPN component of both FortiOS and FortiProxy. By exploiting a buffer overflow in the heap memory, attackers can execute malicious code remotely. This vulnerability is particularly dangerous because it can be triggered without authentication, making it ideal for an initial compromise [5].
CVE-2024-21762
Type: Out-of-Bounds Write in sslvpnd
Impact: Remote Code Execution
This vulnerability affects the SSL-VPN daemon (sslvpnd) in FortiOS. It allows unauthenticated remote attackers to send specially crafted HTTP requests that write data outside of allocated memory bounds. This can lead to arbitrary code execution, giving attackers full control over a device [4].
In short, these flaws enable remote attackers to execute arbitrary code without authentication by exploiting memory corruption issues such as buffer overflows and out-of-bounds writes. Once inside, threat actors use symbolic link (symlink) in order to maintain persistence on target devices across patches and firmware updates. This persistence then enables them to bypass security controls and manipulate firewall configurations, effectively turning patched systems into long-term footholds for deeper network compromise [1][2][3].
Darktrace’s Coverage
Darktrace detected a series of suspicious activities originating from a compromised Fortinet VPN device, including anomalous HTTP traffic, internal network scanning, and SMB reconnaissance, all indicative of post-exploitation behavior. Following initial detection by Darktrace’s real-time models, its Autonomous Response capability swiftly acted on the malicious activity, blocking suspicious connections and containing the threat before further compromise could occur.
Further investigation by Darktrace’s Threat Research team uncovered a stealthy and persistent attack that leveraged known Fortinet SSL-VPN vulnerabilities to facilitate lateral movement and privilege escalation within the network.
The attack on a Darktrace customer likely began on April 11 with the exploitation of a Fortinet VPN device running an outdated version of FortiOS. Darktrace observed a high volume of HTTP traffic originating from this device, specifically targeting internal systems. Notably, many of these requests were directed at the /cgi-bin/ directory, a common target for attackers attempting to exploit web interfaces to run unauthorized scripts or commands. This pattern strongly indicated remote code execution attempts via the SSL-VPN interface [7].
Once access was gained, the threat actor likely modified existing firewall rules, a tactic often used to disable security controls or create hidden backdoors for future access. While Darktrace does not have direct visibility into firewall configuration changes, the surrounding activity and post-exploitation behavior indicated that such modifications were made to support long-term persistence within the network.
Figure 1: HTTP activity from the compromised Fortinet device, including repeated requests to /cgi-bin/ over port 8080
Phase 2: Establishing Persistence & Lateral Movement
Shortly after the initial compromise of the Fortinet VPN device, the threat actor began to expand their foothold within the internal network. Darktrace detected initial signs of network scanning from this device, including the use of Nmap to probe the internal environment, likely in an attempt to identify accessible services and vulnerable systems.
Figure 2: Darktrace’s detection of unusual network scanning activities on the affected device.
Around the same time, Darktrace began detecting anomalous activity on a second device, specifically an internal firewall interface device. This suggested that the attacker had established a secondary foothold and was leveraging it to conduct deeper reconnaissance and move laterally through the network.
In an effort to maintain persistence within the network, the attackers likely deployed symbolic links in the SSL-VPN language file directory on the Fortinet device. While Darktrace did not directly observe symbolic link abuse, Fortinet has identified this as a known persistence technique in similar attacks [2][3]. Based on the observed post-exploitation behavior and likely firewall modifications, it is plausible that such methods were used here.
With lateral movement initiated from the internal firewall interface device, the threat actor proceeded to escalate their efforts to map the internal network and identify opportunities for privilege escalation.
Darktrace observed a successful NTLM authentication from the internal firewall interface to the domain controller over the outdated protocol SMBv1, using the account ‘anonymous’. This was immediately followed by a failed NTLM session connection using the hostname ‘nmap’, further indicating the use of Nmap for enumeration and brute-force attempts. Additional credential probes were also identified around the same time, including attempts using the credential ‘guest’.
Figure 3: Darktrace detection of a series of login attempts using various credentials, with a mix of successful and unsuccessful attempts.
The attacker then initiated DCE_RPC service enumeration, with over 300 requests to the Endpoint Mapper endpoint on the domain controller. This technique is commonly used to discover available services and their bindings, often as a precursor to privilege escalation or remote service manipulation.
Over the next few minutes, Darktrace detected more than 1,700 outbound connections from the internal firewall interface device to one of the customer’s subnets. These targeted common services such as FTP (port 21), SSH (22), Telnet (23), HTTP (80), and HTTPS (443). The threat actor also probed administrative and directory services, including ports 135, 137, 389, and 445, as well as remote access via RDP on port 3389.
Further signs of privilege escalation attempts were observed with the detection of over 300 Netlogon requests to the domain controller. Just over half of these connections were successful, indicating possible brute-force authentication attempts, credential testing, or the use of default or harvested credentials.
Figure 4: Netlogon and DCE-RPC activity from the affected device, showing repeated service bindings to epmapper and Netlogon, followed by successful and failed NetrServerAuthenticate3 attempts.
Phase 4: Privilege Escalation & Remote Access
A few minutes later, the attacker initiated an RDP session from the internal firewall interface device to an internal server. The session lasted over three hours, during which more than 1.5MB of data was uploaded and over 5MB was downloaded.
Notably, no RDP cookie was observed during this session, suggesting manual access, tool-less exploitation, or a deliberate attempt to evade detection. While RDP cookie entries were present on other occasions, none were linked to this specific session—reinforcing the likelihood of stealthy remote access.
Additionally, multiple entries during and after this session show SSL certificate validation failures on port 3389, indicating that the RDP connection may have been established using self-signed or invalid certificates, a common tactic in unauthorized or suspicious remote access scenarios.
Figure 5: Darktrace’s detection of an RDP session from the firewall interface device to the server, lasting over 3 hours.
Darktrace Autonomous Response
Throughout the course of this attack, Darktrace’s Autonomous Response capability was active on the customer’s network. This enabled Darktrace to autonomously intervene by blocking specific connections and ports associated with the suspicious activity, while also enforcing a pre-established “pattern of life” on affected devices to ensure they were able to continue their expected business activities while preventing any deviations from it. These actions were crucial in containing the threat and prevent further lateral movement from the compromised device.
Figure 6: Darktrace’s Autonomous Response targeted specific connections and restricted affected devices to their expected patterns of life.
Conclusion
This incident highlights the importance of important staying on top of patching and closely monitoring VPN infrastructure, especially for internet-facing systems like Fortinet devices. Despite available patches, attackers were still able to exploit known vulnerabilities to gain access, move laterally and maintain persistence within the customer’s network.
Attackers here demonstrated a high level of stealth and persistence. Not only did they gain access to the network and carry out network scans and lateral movement, but they also used techniques such as symbolic link abuse, credential probing, and RDP sessions without cookies to avoid detection. Darktrace’s detection of the post-exploitation activity, combined with the swift action of its Autonomous Response technology, successfully blocked malicious connections and contained the attack before it could escalate
Credit to Priya Thapa (Cyber Analyst), Vivek Rajan (Cyber Analyst), and Ryan Traill (Analyst Content Lead)
The content provided in this blog is published by Darktrace for general informational purposes only and reflects our understanding of cybersecurity topics, trends, incidents, and developments at the time of publication. While we strive to ensure accuracy and relevance, the information is provided “as is” without any representations or warranties, express or implied. Darktrace makes no guarantees regarding the completeness, accuracy, reliability, or timeliness of any information presented and expressly disclaims all warranties.
Nothing in this blog constitutes legal, technical, or professional advice, and readers should consult qualified professionals before acting on any information contained herein. Any references to third-party organizations, technologies, threat actors, or incidents are for informational purposes only and do not imply affiliation, endorsement, or recommendation.
Darktrace, its affiliates, employees, or agents shall not be held liable for any loss, damage, or harm arising from the use of or reliance on the information in this blog.
The cybersecurity landscape evolves rapidly, and blog content may become outdated or superseded. We reserve the right to update, modify, or remove any content without notice.