Blog
/
Identity
/
May 19, 2023

Darktrace Stops Large-Scale Account Hijack

Learn how Darktrace detected and stopped a large-scale account hijack that led to a phishing attack. Protect your business with these insights.
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
Zoe Tilsiter
Cyber Analyst
Default blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog image
19
May 2023

Introduction 

As malicious actors across the threat landscape continue to take advantage of the widespread adoption of Software-as-a-Service (SaaS) platforms and multi-factor authentication (MFA) services to gain unauthorized access to organizations’ networks, it is crucial to have appropriate security tools in place to defend against account compromise at the earliest stage.

One method frequently employed by attackers is account takeover. Account takeovers occur when a threat actor exploits credentials to login to a SaaS account, often from an unusual location where the genuine actor does not usually login from. 

Access to these accounts can be caused by harvesting credentials through phishing emails and password spray attacks, or by exploiting insecure cloud safety practices such as not having MFA enabled on user accounts, requiring only user credentials for authentication. Once the integrity of the account is compromised, the threat actor can conduct further activity, such as delivering malware, reading and exfiltrating sensitive data, and sending out phishing emails to harvest further internal and external user credentials, repeating the attack cycle [1,2]. 

In early 2023, Darktrace detected a large-scale account takeover and phishing attack on the network of a customer in the education sector that affected hundreds of accounts and resulted in thousands of emails being forwarded outside of the network. The exceptional degree of visibility provided by Darktrace DETECT™ allowed for the detection of adversarial activity at every stage of the kill chain, and direct support from the Darktrace Analyst team via the Ask the Expert (ATE) service ensured the customer was fully informed and equipped to implement remedial action. 

Details of Attack Chain

Darktrace observed the same pattern of activity on all hijacked accounts on the customer’s network; login from unfamiliar locations, enablement of a mail forwarding rule that forwards all incoming emails to malicious email addresses, and the sending of phishing emails followed by their deletion. 

Figure 1: Timeline of attack on hijacked SaaS accounts.

Initial Access

Darktrace DETECT first detected anomalous SaaS activity on the customer environment on January 14, 2023, and then again on February 3, when multiple SaaS accounts were observed logging in from atypical locations with rare IP addresses and geographically impossible travel timings, or logging in whilst the account owner was active elsewhere. Subsequent investigation using open-source intelligence (OSINT) sources revealed one of the IP addressed had recently been associated with brute-force or password spray attempt.

This pattern of unusual login behavior persisted throughout the timeframe of the attack, with more unique accounts generating model breaches each day for similarly anomalous logins. As MFA authentication was not enforced for these user logins, the initial intrusion process was enabled by requiring only credentials for authentication.

Sending Emails 

The compromised accounts were also seen sending out emails with the subject ‘Email HELP DESK’ to external and internal recipients. This was likely represented a threat actor employing social engineering tactics to gain the trust of the recipient by posing as an internal help desk.

Mail Forwarding

Following the successful logins, compromised accounts began creating email rules to forward mail to external email addresses, some of which were associated with domains that had hits for malicious activity according to OSINT sources [3].

  • chotunai[.]com
  • bymercy[.]com
  • breazeim[.]com
  • brandoza[.]com

Forwarding mail is a commonly observed tactic during SaaS compromises to control lines of communication. Malicious actors often attempt to insert themselves into ongoing correspondence for illicit purposes, such as exfiltrating sensitive information, gaining persistent access to the compromised email or redirecting invoice payments. 

Email Deletions

Shortly after the mail forwarding activity, compromised accounts were detected performing anomalous email deletions en masse. Further investigation revealed that these accounts had previously sent a large volume of phishing emails and this mass deletion likely represented an attempt to conceal these activities by deleting them from their outboxes.

On February 10, the customer applied a mass password reset on all accounts that Darktrace had identified as compromised and provisioned, privileged accounts with MFA. They have indicated that those measures successfully halted the compromise, addressing the initial point of entry.  

Darktrace Coverage

Using its Self-Learning AI, Darktrace effectively demonstrated its ability to detect unusual SaaS activity that could indicate that an account has been hijacked by malicious actors. Rather than relying on a traditional rules and signature-based approach, Darktrace models develop an understanding of the network itself and can instantly recognize when a compromised deviates from its expected pattern of life.

Figure 2: Detection of unusual SaaS activity on hijacked SaaS account.

Initial Access

Initial access was detected by the following models:

  • Security Integration / High Severity Integration Detection  
  • SaaS / Unusual Activity / Activity from Multiple Unusual IPs 
  • SaaS / Access / Unusual External Source for SaaS Credential Use 
  • SaaS / Compromise / Login From Rare Endpoint While User Is Active 

Initial access was also detected by the following Cyber AI Analyst Incidents:

  • Possible Hijack of Office365 Account 

The model breaches and AI Analyst incidents detected logins from 100% rare external IP addresses in conjunction with a lack of MFA usage, as depicted in Figure 3.

Figure 3: Breach log showing initial detection of a SaaS login from a 100% rare IP where MFA was not used.
Figure 4: Initial detection of unusual SaaS activity visualized in Darktrace's SaaS console.

Mail Forwarding

Mail forwarding was detected by the following models:

  • SaaS / Admin / Mail Forwarding Enabled 

Compromised accounts were largely detected configuring mail forwarding rules to external email addresses, ostensibly to establish persistence on the network and exfiltrate sensitive correspondence.

Figure 5: The enablement of mail forwarding was detected as 100% new or uncommon for the account in question.

Mass Email Deletion

Mass email deletion was detected by the following models:

  • SaaS / Compromise / Suspicious Login and Mass Email Deletes 
  • SaaS / Resource / Mass Email Deletes from Rare Location 
Figure 6: Compromised account deleting phishing emails it had previously sent from the outbox.

Darktrace detected accounts performing highly anomalous mass email deletions from rare locations. The actors deleted the email “Email HELP DESK” which was later confirmed as being the primary phishing email used in the attack. Deletions were observed on compromised accounts’ outboxes, presumably to conceal the malicious activity.

Darktrace also detected this linked pattern of activity in sequential models such as: 

  • SaaS / Compromise / Unusual Login, Sent Mail, Deleted Sent
  • SaaS / Compromise / Suspicious Login and Mass Email Deletes 

Ask the Expert

The customer used the ATE service to request more technical information and support concerning the attack. Darktrace’s 24/7 team of analysts were able to offer expert assistance and further details to assist in the subsequent investigations and remediation steps. 

Further Detection and Response  

Unfortunately, the customer did not have Darktrace/Email™ enabled at the time of the attack. Darktrace/Email has visibility over inbound and outbound mail-flow which provides an oversight on potential data loss incidents. In this case, Darktrace DETECT/Email would have been able to provide full visibility over the phishing emails sent by the compromised accounts, as well as the attackers attempts to spoof an internal helpdesk. Further to this, the new Analysis Outlook integration helps employees understand why an email is suspicious and enables them report emails directly to the security team, which helps to continuously build user awareness of phishing attacks. 

Darktrace/Email also enhances Darktrace/Network™ detections by triggering ‘Email Nexus’ models within Darktrace/Network, where malicious activity is detected across the digital estate, correlating moving from SaaS compromised logins to mass email spam being sent out by compromised users

Figure 7: Email Nexus models within the Darktrace/Network enhanced by Darktrace/Email

Darktrace RESPOND™ was not enabled on the customer environment at the time of the attack; if it were, Darktrace would have been able to autonomously take action against the SaaS model breaches detecting across multiple of the kill chain. RESPOND would have disabled the hijacked accounts or force them to log out for a period of time, whilst also disabling the inbox rules that had been established by malicious actors. This would have given the customer’s security team valuable time to analyze the incident and mitigate the situation, preventing the attack from escalating any further. 

Conclusion

Ultimately, Darktrace demonstrated its unparalleled visibility over customer networks which allowed for the detection of this large-scale targeted SaaS account takeover, and the subsequent phishing attack. It underscores the importance of defense in depth; critically, MFA was not enforced for this environment which likely made the targeted organization far more susceptible to compromise via credential theft. The phishing activity detected by Darktrace following this account compromise also highlights the need for email protection in any security stack. 

Darktrace’s visibility meant allowed it to detect the attack at a high degree of granularity, including the account logins, email forwarding rule creations, outbound mail, and the mass deletions of phishing emails. Darktrace’s anomaly-based detection means it does not have to rely on signatures, rules or known indicators of compromise (IoCs) when identifying an emerging threat, instead placing the emphasis on recognizing a user’s deviation from its normal behavior.

However, without the presence of an autonomous response technology able to instantly intervene and stop ongoing attacks, organizations will always be reacting to attacks once the damage is done. Darktrace RESPOND is uniquely placed to take action against suspicious activity as soon as it is detected, preventing attacks from escalating and saving customers from significant disruption to their business.

Credit to: Zoe Tilsiter, Cyber Analyst, Gernice Lee, Cyber Analyst.

Appendices

Models Breached

SaaS / Access / Unusual External Source for SaaS Credential Use

SaaS / Admin / Mail Forwarding Enabled

SaaS / Compliance / Microsoft Cloud App Security Alert Detected

SaaS / Compromise / SaaS Anomaly Following Anomalous Login 

SaaS / Compromise / Unusual Login, Sent Mail, Deleted Sent

SaaS / Compromise / Suspicious Login and Mass Email Deletes 

SaaS / Resource / Mass Email Deletes from Rare Location

SaaS / Unusual Activity / Multiple Unusual External Sources For SaaS Credential

SaaS / Unusual Activity / Activity from Multiple Unusual IPs

SaaS / Unusual Activity / Multiple Unusual SaaS Activities 

Security Integration / Low Severity Integration Detection

Security Integration / High Severity Integration Detection

List of IoCs

brandoza[.]com - domain - probable domain of forwarded email address

breazeim[.]com - domain - probable domain of forwarded email address

bymercy[.]com - domain - probable domain of forwarded email address

chotunai[.]com - domain - probable domain of forwarded email address

MITRE ATT&CK Mapping

Tactic: INITIAL ACCESS, PERSISTENCE, PRIVILEGE ESCILATION, DEFENSE EVASION

Technique: T1078.004 – Cloud Accounts

Tactic: COLLECTION

Technique: T1114- Email Collection

Tactic:COLLECTION

Technique: T1114.003- Email Forwarding Rule

Tactic: IMPACT

Technique: T1485- Data Destruction

Tactic: DEFENSE EVASION

Technique: T1578.003 – Delete Cloud Instance

References

[1] Darktrace, 2022, Cloud Application Security_ Protect your SaaS with Self-Learning AI.pdf

[2] https://www.cloudflare.com/en-gb/learning/access-management/account-takeover/ 

[3] https://www.virustotal.com/gui/domain/chotunai.com 

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
Zoe Tilsiter
Cyber Analyst

More in this series

No items found.

Blog

/

Network

/

March 30, 2026

Phantom Footprints: Tracking GhostSocks Malware

ghostsocks malwareDefault blog imageDefault blog image

Why are attackers using residential proxies?

In today's threat landscape, blending in to normal activity is the key to success for attackers and the growing reliance on residential proxies shows a significant shift in how threat actors are attempting to bypass IP detection tools.

The increasing dependency on residential proxies has exposed how prevalent proxy services are and how reliant a diverse range of threat actors are on them. From cybercriminal groups to state‑sponsored actors, the need to bypass IP detection tools is fundamental to the success of these groups. One malware that has quietly become notorious for its ability to avoid anomaly detection is GhostSocks, a malware that turns compromised devices into residential proxies.

What is GhostSocks?

Originally marketed on the Russian underground forum xss[.]is as a Malware‑as‑a‑Service (MaaS), GhostSocks enables threat actors to turn compromised devices into residential proxies, leveraging the victim's internet bandwidth to route malicious traffic through it.

How does Ghostsocks malware work? 

The malware offers the threat actor a “clean” IP address, making it look like it is coming from a household user. This enables the bypassing of geographic restrictions and IP detection tools, a perfect tool for avoiding anomaly detection. It wasn’t until 2024, when a partnership was announced with the infamous information stealer Lumma Stealer, that GhostSocks surged into widespread adoption and alluded to who may be the author of the proxy malware.

Written in GoLang, GhostSocks utilizes the SOCKS5 proxy protocol, creating a SOCKS5 connection on infected devices. It uses a relay‑based C2 implementation, where an intermediary server sits in between the real command-and-control (C2) server and the infected device.

How does Ghostsocks malware evade detection?

To further increase evasion, the Ghostsocks malware wraps its SOCKS5 tunnels in TLS encryption, allowing its malicious traffic to blend into normal network traffic.

Early variants of GhostSocks do not implement a persistence mechanism; however, later versions achieve persistence via registry run keys, ensuring sustained proxy operational time [1].

While proxying is its primary purpose, GhostSocks also incorporates backdoor functionality, enabling malicious actors to run arbitrary commands and download and deploy additional malicious payloads. This was evident with the well‑known ransomware group Black Basta, which reportedly used GhostSocks as a way of maintaining long‑term access to victims’ networks [1].

Darktrace’s detection of GhostSocks Malware

Darktrace observed a steady increase in GhostSocks activity across its customer base from late 2025, with its Threat Research team identifying multiple incidents involving the malware. In one notable case from December 2025, Darktrace detected GhostSocks operating alongside Lumma Stealer, reinforcing that the partnership between Lumma and GhostSocks remains active despite recent attempts to disrupt Lumma’s infrastructure.

Darktrace’s first detection of GhostSocks‑related activity came when a device on the network of a customer in the education sector began making connections to an endpoint with a suspicious self‑signed certificate that had never been seen on the network before.

The endpoint in question, 159.89.46[.]92 with the hostname retreaw[.]click, has been flagged by multiple open‑source intelligence (OSINT) sources as being associated with Lumma Stealer’s C2 infrastructure [2], indicating its likely role in the delivery of malicious payloads.

Darktrace’s detection of suspicious SSL connections to retreaw[.]click, indicating an attempted link to Lumma C2 infrastructure.
Figure 1: Darktrace’s detection of suspicious SSL connections to retreaw[.]click, indicating an attempted link to Lumma C2 infrastructure.

Less than two minutes later, Darktrace observed the same device downloading the executable (.exe) file “Renewable.exe” from the IP 86.54.24[.]29, which Darktrace recognized as 100% rare for this network.

Darktrace’s detection of a device downloading the unusual executable file “Renewable.exe”.
Figure 2: Darktrace’s detection of a device downloading the unusual executable file “Renewable.exe”.

Both the file MD5 hash and the executable itself have been identified by multiple OSINT vendors as being associated with the GhostSocks malware [3], with the executable likely the backdoor component of the GhostSocks malware, facilitating the distribution of additional malicious payloads [4].

Following this detection, Darktrace’s Autonomous Response capability recommended a blocking action for the device in an early attempt to stop the malicious file download. In this instance, Darktrace was configured in Human Confirmation Mode, meaning the customer’s security team was required to manually apply any mitigative response actions. Had Autonomous Response been fully enabled at the time of the attack, the connections to 86.54.24[.]29 would have been blocked, rendering the malware ineffective at reaching its C2 infrastructure and halting any further malicious communication.

 Darktrace’s Autonomous Response capability suggesting blocking the suspicious connections to the unusual endpoint from which the malicious executable was downloaded.
Figure 3: Darktrace’s Autonomous Response capability suggesting blocking the suspicious connections to the unusual endpoint from which the malicious executable was downloaded.

As the attack was able to progress, two days later the device was detected downloading additional payloads from the endpoint www.lbfs[.]site (23.106.58[.]48), including “Setup.exe”, “,.exe”, and “/vp6c63yoz.exe”.

Darktrace’s detection of a malicious payload being downloaded from the endpoint www.lbfs[.]site.
Figure 4: Darktrace’s detection of a malicious payload being downloaded from the endpoint www.lbfs[.]site.

Once again, Darktrace recognized the anomalous nature of these downloads and suggested that a “group pattern of life” be enforced on the offending device in an attempt to contain the activity. By enforcing a pattern of life on a device, Darktrace restricts its activity to connections and behaviors similar to those performed by peer devices within the same group, while still allowing it to carry out its expected activity, effectively preventing deviations indicative of compromise while minimizing disruption. As mentioned earlier, these mitigative actions required manual implementation, so the activity was able to continue. Darktrace proceeded to suggest further actions to contain subsequent malicious downloads, including an attempt to block all outbound traffic to stop the attack from progressing.

An overview of download activity and the Autonomous Response actions recommended by Darktrace to block the downloads.
Figure 5: An overview of download activity and the Autonomous Response actions recommended by Darktrace to block the downloads.

Around the same time, a third executable download was detected, this time from the hostname hxxp[://]d2ihv8ymzp14lr.cloudfront.net/2021-08-19/udppump[.]exe, along with the file “udppump.exe”.While GhostSocks may have been present only to facilitate the delivery of additional payloads, there is no indication that these CloudFront endpoints or files are functionally linked to GhostSocks. Rather, the evidence points to broader malicious file‑download activity.

Shortly after the multiple executable files had been downloaded, Darktrace observed the device initiating a series of repeated successful connections to several rare external endpoints, behavior consistent with early-stage C2 beaconing activity.

Cyber AI Analyst’s investigation

Darktrace’s detection of additional malicious file downloads from malicious CloudFront endpoints.
Figure 7: Darktrace’s detection of additional malicious file downloads from malicious CloudFront endpoints.

Throughout the course of this attack, Darktrace’s Cyber AI Analyst carried out its own autonomous investigation, piecing together seemingly separate events into one wider incident encompassing the first suspicious downloads beginning on December 4, the unusual connectivity to many suspicious IPs that followed, and the successful beaconing activity observed two days later. By analyzing these events in real-time and viewing them as part of the bigger picture, Cyber AI Analyst was able to construct an in‑depth breakdown of the attack to aid the customer’s investigation and remediation efforts.

Cyber AI Analyst investigation detailing the sequence of events on the compromised device, highlighting its extensive connectivity to rare endpoints, the related malicious file‑download activity, and finally the emergence of C2 beaconing behavior.
Figure 8: Cyber AI Analyst investigation detailing the sequence of events on the compromised device, highlighting its extensive connectivity to rare endpoints, the related malicious file‑download activity, and finally the emergence of C2 beaconing behavior.

Conclusion

The versatility offered by GhostSocks is far from new, but its ability to convert compromised devices into residential proxy nodes, while enabling long‑term, covert network access—illustrates how threat actors continue to maximise the value of their victims’ infrastructure. Its growing popularity, coupled with its ongoing partnership with Lumma, demonstrates that infrastructure takedowns alone are insufficient; as long as threat actors remain committed to maintaining anonymity and can rapidly rebuild their ecosystems, related malware activity is likely to persist in some form.

Credit to Isabel Evans (Cyber Analyst), Gernice Lee (Associate Principal Analyst & Regional Consultancy Lead – APJ)
Edited by Ryan Traill (Content Manager)

Appendices

References

1.    https://bloo.io/research/malware/ghostsocks

2.    https://www.virustotal.com/gui/domain/retreaw.click/community

3.    https://synthient.com/blog/ghostsocks-from-initial-access-to-residential-proxy

4.    https://www.joesandbox.com/analysis/1810568/0/html

5. https://www.virustotal.com/gui/url/fab6525bf6e77249b74736cb74501a9491109dc7950688b3ae898354eb920413

Darktrace Model Detections

Real-time Detection Models

Anomalous Connection / Suspicious Self-Signed SSL

Anomalous Connection / Rare External SSL Self-Signed

Anomalous File / EXE from Rare External Location

Anomalous File / Multiple EXE from Rare External Locations

Compromise / Possible Fast Flux C2 Activity

Compromise / Large Number of Suspicious Successful Connections

Compromise / Large Number of Suspicious Failed Connections

Compromise / Sustained SSL or HTTP Increase

Autonomous Response Models

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

Antigena / Network / External Threat / Antigena Suspicious File Block

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

Antigena / Network / External Threat / Antigena File then New Outbound Block

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

Antigena / Network / External Threat / Antigena Suspicious Activity Block

MITRE ATT&CK Mapping

Tactic – Technique – Sub-Technique

Resource Development – T1588 - Malware

Initial Access - T1189 - Drive-by Compromise

Persistence – T1112 – Modify Registry

Command and Control – T1071 – Application Layer Protocol

Command and Control – T1095 – Non-application Layer Protocol

Command and Control – T1071 – Web Protocols

Command and Control – T1571 – Non-Standard Port

Command and Control – T1102 – One-Way Communication

List of Indicators of Compromise (IoCs)

86.54.24[.]29 - IP - Likely GhostSocks C2

http[://]86.54.24[.]29/Renewable[.]exe - Hostname - GhostSocks Distribution Endpoint

http[://]d2ihv8ymzp14lr.cloudfront[.]net/2021-08-19/udppump[.]exe - CDN - Payload Distribution Endpoint

www.lbfs[.]site - Hostname - Likely C2 Endpoint

retreaw[.]click - Hostname - Lumma C2 Endpoint

alltipi[.]com - Hostname - Possible C2 Endpoint

w2.bruggebogeyed[.]site - Hostname - Possible C2 Endpoint

9b90c62299d4bed2e0752e2e1fc777ac50308534 - SHA1 file hash – Likely GhostSocks payload

3d9d7a7905e46a3e39a45405cb010c1baa735f9e - SHA1 file hash - Likely follow-up payload

10f928e00a1ed0181992a1e4771673566a02f4e3 - SHA1 file hash - Likely follow-up payload

Continue reading
About the author
Isabel Evans
Cyber Analyst

Blog

/

AI

/

March 27, 2026

State of AI Cybersecurity 2026: 92% of security professionals concerned about the impact of AI agents

Default blog imageDefault blog image

The findings in this blog are taken from Darktrace's annual State of AI Cybersecurity Report 2026.

AI is already embedded in day-to-day enterprise activity, with 78% of participants in one recent survey reporting that their organizations are using generative AI in at least one business function. Generative AI now acts as an always-on assistant, researcher, creator, and coach across an expanding array of departments and functions. Autonomous agents are performing multi-step operational workflows from end to end. AI features have been layered on top of every SaaS application. And vibe coding is making it possible for employees without deep technical expertise to build their own AI-powered automations.

According to Gartner, more than 80% of enterprises will have deployed GenAI models, applications, or APIs in production environments by the end of this year, up from less than 5% in 2023. Companies report a 130% increase in spending on AI over the same period, with 72% of business leaders using AI tools at least weekly. The outsized efficiency and productivity gains that were once a future vision are quickly becoming everyday reality.

AI is currently driving business growth and innovation, and organizations risk falling behind peers if they don’t keep up with the pace of adoption, but it is also quietly expanding the enterprise attack surface. The modern CISO is challenged to both enable innovation and protect the business from these emerging threats.

AI agents introduce new risks and vulnerabilities

AI agents are playing growing roles in enterprise production environments. In many cases, these agents act with broad permissions across multiple software systems and platforms. This means they’re granted far-reaching access – to sensitive data, business-critical applications, tokens and APIs, and IT and security tools. With this access comes risk for security leaders – 92% are concerned about the use of AI agents across the workforce and their impact on security.

These agents must be governed as identities, with least-privilege access and ongoing monitoring. They can’t be thought of as invisible aspects of the application estate. Understanding how AI agents behave, and how to manage their permissions, control their behavior, and limit their data access will be a top security priority throughout 2026.

Generative AI prompts: The next frontier

Prompts are how users – both human and agentic – interact with AI systems, and they’re where natural language gets translated into model behavior. Natural language is infinite in its potential combinations and permutations, making this aspect of the attack surface open-ended and far more complex than traditional CVEs. With carefully crafted prompts, bad actors may be able to coax models into disclosing sensitive data, bypassing guardrails, or initiating undesirable actions.

Among security leaders, the biggest worries about AI usage in their environments all involve ways that systems might be manipulated to bypass traditional controls.

  • 61% are most concerned about the exposure of sensitive data
  • 56% are most concerned about potential data security and policy violations
  • 51% are most concerned about the misuse or abuse of AI tools

The more employees rely on AI in their day-to-day workflows, the more critical it becomes for security teams to understand how prompt behavior determines model behavior – and where that behavior could go wrong.

What does “securing AI” mean in practice?

AI adoption opens new security risks that blur the boundaries between traditional security disciplines. A single malicious interaction with an AI model could involve identity misuse, sensitive data exposure, application logic abuse, and supply chain risk – all within a single workflow. Protecting this dynamic and rapidly evolving attack surface requires an approach that spans identity security, cloud security, application security, data security, software development security, and more.

The task for security leaders is to implement the tools, policies, and frameworks to mitigate these novel, expansive, and cross-disciplinary risks.

However, within most enterprises, AI policy creation remains in its infancy. Just 37% of security leaders report that their organization has a formal AI policy, representing a small but worrisome decrease from last year. Conversations about AI abound: in 52% of organizations, there’s discussion about an AI policy. Still, talk is cheap, and leaders will need to take action if they’re to successfully enable secure AI innovation.

To govern and protect their AI systems, organizations must take a multi-pronged approach. This requires building out policies, but it also demands that they are able to:

  • Monitor the prompts driving GenAI assistants and agents in real time. Organizations must be able to inspect prompts, sessions, and responses across enterprise GenAI tools, low- and high-code environments, and SaaS and SASE so that they can detect clever conversational prompt attacks and malicious chaining.
  • Secure all business AI agent identities. Security teams need to identify all the agents acting within their environment and supply chain, map their connections and interactions via MCP and services like Amazon S3, and audit their behavior across the cloud, SaaS environments, and on the network and endpoint devices.
  • Maintain centralized, comprehensive visibility. Understanding intent, assessing risks, and enforcing policies all require that security teams have a single view that spans AI interactions across the entire business.
  • Discover and control shadow AI. Teams need to be able to identify unsanctioned AI activities, distinguish the misuse of legitimate tools from their appropriate use, and apply policies to protect data, while guiding users towards approved solutions.

Scaling AI safely and responsibly

The approach that most cybersecurity vendors have taken – using historical patterns to predict future threats – doesn’t work well for AI systems. Because AI changes its behavior in response to the information it encounters while taking action, previous patterns don’t indicate what it will do next. Looking at past attacks can’t tell you how complex models will behave in your individual business.

Securing AI requires interpreting ambiguous interactions, uncovering subtleties that reveal intent within extended conversations, understanding how access accumulates over time, and recognizing when behavior – both human and machine – begins to drift towards areas of risk. To do this, you need to understand what “normal” looks like in each unique organization: how users, systems, applications, and AI agents behave, how they communicate, and how data flows between them.

Darktrace has spent more than a decade designing AI-powered solutions that can understand and adapt to evolving behavior in complex environments. This technology learns directly from the environment it protects, identifying malicious actions that deviate from normal operations, so that it can stop AI-related threats on the very first encounter.

As AI adoption reshapes enterprise operations, humans and machines will collaborate more and more often. This collaboration might dramatically expand the attack surface, but it also has the potential to be a force multiplier for defenders.

Explore the full State of AI Cybersecurity 2026 report for deeper insights into how security leaders are responding to AI-driven risks.

Learn more about securing AI in your enterprise.

[related-resource]

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