Blog
/
Network
/
December 9, 2024

Darktrace’s view on Operation Lunar Peek: Exploitation of Palo Alto firewall devices (CVE 2024-0012 and 2024-9474)

Darktrace’s Threat Research team investigated a major campaign exploiting vulnerabilities in Palo Alto firewall devices (CVE 2024-0012 and 2024-9474). Learn about the spike in post-exploitation activities and understand the need for anomaly-based detection to stay ahead of evolving threats.
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
Adam Potter
Senior Cyber Analyst
Default blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog imageDefault blog image
09
Dec 2024

Introduction: Spike in exploitation and post-exploitation activity affecting Palo Alto firewall devices

As the first line of defense for many organizations, perimeter devices such as firewalls are frequently targeted by threat actors. If compromised, these devices can serve as the initial point of entry to the network, providing access to vulnerable internal resources. This pattern of malicious behavior has become readily apparent within the Darktrace customer base. In 2024, Darktrace Threat Research analysts identified and investigated at least two major campaigns targeting internet-exposed perimeter devices. These included the exploitation of PAN-OS firewall exploitation via CVE 2024-3400 and FortiManager appliances via CVE 2024-47575.

More recently, at the end of November, Darktrace analysts observed a spike in exploitation and post-exploitation activity affecting, once again, Palo Alto firewall devices in the days following the disclosure of the CVE 2024-0012 and CVE-2024-9474 vulnerabilities.

Threat Research analysts had already been investigating potential exploitation of the firewalls’ management interface after Palo Alto published a security advisory (PAN-SA-2024-0015) on November 8. Subsequent analysis of data from Darktrace’s Security Operations Center (SOC) and external research uncovered multiple cases of Palo Alto firewalls being targeted via the likely exploitation of these vulnerabilities since November 13, through the end of the month. Although this spike in anomalous behavior may not be attributable to a single malicious actor, Darktrace Threat Research identified a clear increase in PAN-OS exploitation across the customer base by threat actors likely utilizing the recently disclosed vulnerabilities, resulting in broad patterns of post-exploitation activity.

How did exploitation occur?

CVE 2024-0012 is an authentication bypass vulnerability affecting unpatched versions of Palo Alto Networks Next-Generation Firewalls. The vulnerability resides in the management interface application on the firewalls specifically, which is written in PHP. When attempting to access highly privileged scripts, users are typically redirected to a login page. However, this can be bypassed by supplying an HTTP request where a Palo Alto related authentication header can be set to “off”.  Users can supply this header value to the Nginx reverse proxy server fronting the application which will then send it without any prior processing [1].

CVE-2024-9474 is a privilege escalation vulnerability that allows a PAN-OS administrator with access to the management web interface to execute root-level commands, granting full control over the affected device [2]. When combined, these vulnerabilities enable unauthenticated adversaries to execute arbitrary commands on the firewall with root privileges.

Post-Exploitation Patterns of Activity

Darktrace Threat Research analysts examined potential indicators of PAN-OS software exploitation via CVE 2024-0012 and CVE-2024-9474 during November 2024. The investigation identified three main groupings of post-exploitation activity:

  1. Exploit validation and initial payload retrieval
  2. Command and control (C2) connectivity, potentially featuring further binary downloads
  3. Potential reconnaissance and cryptomining activity

Exploit Validation

Across multiple investigated customers, Darktrace analysts identified likely vulnerable PAN-OS devices conducting external network connectivity to bin services. Specifically, several hosts performed DNS queries for, and HTTP requests to Out-of-Band Application Security Testing (OAST) domains, such as csv2im6eq58ujueonqs0iyq7dqpak311i.oast[.]pro. These endpoints are commonly used by network administrators to harden defenses, but they are increasingly used by threat actors to verify successful exploitation of targeted devices and assess their potential for further compromise. Although connectivity involving OAST domains were prevalent across investigated incidents, this activity was not necessarily the first indicator observed. In some cases, device behavior involving OAST domains also occurred shortly after an initial payload was downloaded.

Darktrace model alert logs detailing the HTTP request to an OAST domain immediately following PAN-OS device compromise.
Figure 1: Darktrace model alert logs detailing the HTTP request to an OAST domain immediately following PAN-OS device compromise.

Initial Payload Retrieval

Following successful exploitation, affected devices commonly performed behaviors indicative of initial payload download, likely in response to incoming remote command execution. Typically, the affected PAN-OS host would utilize the command line utilities curl and Wget, seen via use of user agents curl/7.61.1 and Wget/1.19.5 (linux-gnu), respectively.

In some cases, the use of these command line utilities by the infected devices was considered new behavior. Given the nature of the user agents, interaction with the host shell suggests remote command execution to achieve the outgoing payload requests.

While additional binaries and scripts were retrieved in later stages of the post-exploitation activity in some cases, this set of behaviors and payloads likely represent initial persistence and execution mechanisms that will enable additional functionality later in the kill chain. During the investigation, Darktrace analysts noted the prevalence of shell script payload requests. Devices analyzed would frequently make HTTP requests over the usual destination port 80 using the command line URL utility (curl), as seen in the user-agent field.

The observed URIs often featured requests for text files, such as “1.txt”, or shell scripts such as “y.sh”. Although packet capture (PCAP) samples were unavailable for review, external researchers have noted that the IP address hosting such “1.txt” files (46.8.226[.]75) serves malicious PHP payloads. When examining the contents of the “y.sh” shell script, Darktrace analysts noticed the execution of bash commands to upload a PHP-written web shell on the affected server.

PCAP showing the client request and server response associated with the download of the y.sh script from 45.76.141[.]166. The body content of the HTTP response highlights a shebang command to run subsequent code as bash script. The content is base64 encoded and details PHP script for what appears to be a webshell that will likely be written to the firewall device.
Figure 2: PCAP showing the client request and server response associated with the download of the y.sh script from 45.76.141[.]166. The body content of the HTTP response highlights a shebang command to run subsequent code as bash script. The content is base64 encoded and details PHP script for what appears to be a webshell that will likely be written to the firewall device.

While not all investigated cases saw initial shell script retrieval, affected systems would commonly make an external HTTP connection, almost always via Wget, for the Executable and Linkable Format (ELF) file “/palofd” from the rare external IP  38.180.147[.]18.

Such requests were frequently made without prior hostname lookups, suggesting that the process or script initiating the requests already contained the external IP address. Analysts noticed a consistent SHA1 hash present for all identified instances of “/palofd” downloads (90f6890fa94b25fbf4d5c49f1ea354a023e06510). Multiple open-source intelligence (OSINT) vendors have associated this hash sample with Spectre RAT, a remote access trojan with capabilities including remote command execution, payload delivery, process manipulation, file transfers, and data theft [3][4].

Figure 3: Advanced Search log metrics highlighting details of the “/palofd” file download over HTTP.

Several targeted customer devices were observed initiating TLS/SSL connections to rare external IPs with self-signed TLS certificates following exploitation. Model data from across the Darktrace fleet indicated some overlap in JA3 fingerprints utilized by affected PAN-OS devices engaging in the suspicious TLS activity. Although JA3 hashes alone cannot be used for process attribution, this evidence suggests some correlation of source process across instances of PAN-OS exploitation.

These TLS/SSL sessions were typically established without the specification of a Server Name Indication (SNI) within the TLS extensions. The SNI extension prevents servers from supplying an incorrect certificate to the requesting client when multiple sites are hosted on the same IP. SSL connectivity without SNI specification suggests a potentially malicious running process as most software establishing TLS sessions typically supply this information during the handshake. Although the encrypted nature of the connection prevented further analysis of the payload packets, external sources note that JavaScript content is transmitted during these sessions, serving as initial payloads for the Sliver C2 platform using Wget [5].

C2 Communication and Additional Payloads

Following validation and preliminary post-compromise actions, examined hosts would commonly initiate varying forms of C2 connectivity. During this time, devices were frequently detected making further payload downloads, likely in response to directives set within C2 communications.

Palo Alto firewalls likely exploited via the newly disclosed CVEs would commonly utilize the Sliver C2 platform for external communication. Sliver’s functionality allows for different styles and formatting for communication. An open-source alternative to Cobalt Strike, this framework has been increasingly popular among threat actors, enabling the generation of dynamic payloads (“slivers”) for multiple platforms, including Windows, MacOS, Linux.

These payloads allow operators to establish persistence, spawn new shells, and exfiltrate data. URI patterns and PCAPs analysis yielded evidence of both English word type encoding within Sliver and Gzip formatting.

For example, multiple devices contacted the Sliver-linked IP address 77.221.158[.]154 using HTTP to retrieve Gzip files. The URIs present for these requests follow known Sliver Gzip formatted communication patterns [6]. Investigations yielded evidence of both English word encoding within Sliver, identified through PCAP analysis, and Gzip formatting.

Sample of URIs observed in Advanced Searchhighlighting HTTP requests to 77.221.158[.]154 for Gzip content suggest of Sliver communication.
Figure 4: Sample of URIs observed in Advanced Searchhighlighting HTTP requests to 77.221.158[.]154 for Gzip content suggest of Sliver communication.
PCAP showing English word encoding for Sliver communication observed during post-exploitation C2 activity.
Figure 5: PCAP showing English word encoding for Sliver communication observed during post-exploitation C2 activity.

External connectivity during this phase also featured TCP connection attempts over uncommon ports for common application protocols. For both Sliver and non-Sliver related IP addresses, devices utilized destination ports such as 8089, 3939, 8880, 8084, and 9999 for the HTTP protocol. The use of uncommon destination ports may represent attempts to avoid detection of connectivity to rare external endpoints. Moreover, some external beaconing within included URIs referencing the likely IP of the affected device. Such behavior can suggest the registration of compromised devices with command servers.

Targeted devices also proceeded to download additional payloads from rare external endpoints as beaconing/C2 activity was ongoing. For example, the newly registered domain repositorylinux[.]org (IP: 103.217.145[.]112) received numerous HTTP GET requests from investigated devices throughout the investigation period for script files including “linux.sh” and “cron.sh”. Young domains, especially those that present as similar to known code repositories, tend to host harmful content. Packet captures of the cron.sh file reveal commands within the HTTP body content involving crontab operations, likely to schedule future downloads. Some hosts that engaged in connectivity to the fake repository domain were later seen conducting crypto-mining connections, potentially highlighting the download of miner applications from the domain.

Additional payloads observed during this time largely featured variations of shell scripts, PHP content, and/or executables. Typically, shell scripts direct the device to retrieve additional content from external servers or repositories or contain potential configuration details for subsequent binaries to run on the device. For example, the “service.sh” retrieves a tar-compressed archive, a configuration JSON file as well as a file with the name “solr” from GitHub, potentially associated with the Apache Solr tool used for enterprise search. These could be used for further enumeration of the host and/or the network environment. PHP scripts observed may involve similar web shell functionality and were retrieved from both rare external IPs identified as well by external researchers [7]. Darktrace also detected the download of octet-stream data occurring mid-compromise from an Amazon Web Services (AWS) S3 bucket. Although no outside research confirmed the functionality, additional executable downloads for files such as “/initd”(IP: 178.215.224[.]246) and “/x6” (IP: 223.165.4[.]175) may relate to tool ingress, further Trojan/backdoor functionality, or cryptocurrency mining.

Figure 7: PCAP specifying the HTTP response headers and body content for the service.sh file request. The body content shown includes variable declarations for URLs that will eventually be called by the device shell via bash command.

Reconnaissance and Cryptomining

Darktrace analysts also noticed additional elements of kill chain operations from affected devices after periods of initial exploit activity. Several devices initiated TCP connections to endpoints affiliated with cryptomining pools such as us[.]zephyr[.]herominers[.]com and  xmrig[.]com. Connectivity to these domains indicates likely successful installation of mining software during earlier stages of post-compromise activity. In a small number of instances, Darktrace observed reconnaissance and lateral movement within the time range of PAN-OS exploitation. Firewalls conducted large numbers of internal connectivity attempts across several critical ports related to privileged protocols, including SMB and SSH. Darktrace detected anonymous NTLM login attempts and new usage of potential PAN-related credentials. These behaviors likely constitute attempts at lateral movement to adjacent devices to further extend network compromise impact.

Model alert connection logs detailing the uncommon failed NTLM logins using an anonymous user account following PAN-OS exploitation.
Figure 8: Model alert connection logs detailing the uncommon failed NTLM logins using an anonymous user account following PAN-OS exploitation.

Conclusion

Darktrace Threat Research and SOC analysts increasingly detect spikes in malicious activity on internet-facing devices in the days following the publication of new vulnerabilities. The latest iteration of this trend highlighted how threat actors quickly exploited Palo Alto firewall using authentication bypass and remote command execution vulnerabilities to enable device compromise. A review of the post-exploitation activity during these events reveals consistent patterns of perimeter device exploitation, but also some distinct variations.

Prior campaigns targeting perimeter devices featured activity largely confined to the exfiltration of configuration data and some initial payload retrieval. Within the current campaign, analysts identified a broader scope post-compromise activity consisting not only of payloads downloads but also extensive C2 activity, reconnaissance, and coin mining operations. While the use of command line tools like curl featured prominently in prior investigations, devices were seen retrieving a generally wider array of payloads during the latest round of activity. The use of the Sliver C2 platform further differentiates the latest round of PAN-OS compromises, with evidence of Sliver activity in about half of the investigated cases.

Several of the endpoints contacted by the infected firewall devices did not have any OSINT associated with them at the time of the attack. However, these indicators were noted as unusual for the devices according to Darktrace based on normal network traffic patterns. This reality further highlights the need for anomaly-based detection that does not rely necessarily on known indicators of compromise (IoCs) associated with CVE exploitation for detection. Darktrace’s experience in 2024 of multiple rounds of perimeter device exploitation may foreshadow future increases in these types of comprise operations.  

Credit to Adam Potter (Senior Cyber Analyst), Alexandra Sentenac (Senior Cyber Analyst), Emma Foulger (Principal Cyber Analyst) and the Darktrace Threat Research team.

Get the latest insights on emerging cyber threats

Attackers are adapting, are you ready? This report explores the latest trends shaping the cybersecurity landscape and what defenders need to know in 2025.

  • Identity-based attacks: How attackers are bypassing traditional defenses
  • Zero-day exploitation: The rise of previously unknown vulnerabilities
  • AI-driven threats: How adversaries are leveraging AI to outmaneuver security controls

Stay ahead of evolving threats with expert analysis from Darktrace. Download the report here.

References

[1]: https://labs.watchtowr.com/pots-and-pans-aka-an-sslvpn-palo-alto-pan-os-cve-2024-0012-and-cve-2024-9474/

[2]: https://security.paloaltonetworks.com/CVE-2024-9474

[3]: https://threatfox.abuse[.]ch/ioc/1346254/

[4]:https://www.virustotal.com/gui/file/4911396d80baff80826b96d6ea7e54758847c93fdbcd3b86b00946cfd7d1145b/detection

[5]: https://arcticwolf.com/resources/blog/arctic-wolf-observes-threat-campaign-targeting-palo-alto-networks-firewall-devices/

[6] https://www.immersivelabs.com/blog/detecting-and-decrypting-sliver-c2-a-threat-hunters-guide

[7] https://arcticwolf.com/resources/blog/arctic-wolf-observes-threat-campaign-targeting-palo-alto-networks-firewall-devices/

Appendices

Darktrace Model Alerts

Anomalous Connection / Anomalous SSL without SNI to New External

Anomalous Connection / Application Protocol on Uncommon Port  

Anomalous Connection / Multiple Failed Connections to Rare Endpoint

Anomalous Connection / Multiple HTTP POSTs to Rare Hostname

Anomalous Connection / New User Agent to IP Without Hostname

Anomalous Connection / Posting HTTP to IP Without Hostname

Anomalous Connection / Rare External SSL Self-Signed

Anomalous File / EXE from Rare External Location

Anomalous File / Incoming ELF File

Anomalous File / Mismatched MIME Type From Rare Endpoint

Anomalous File / Multiple EXE from Rare External Locations

Anomalous File / New User Agent Followed By Numeric File Download

Anomalous File / Script from Rare External Location

Anomalous File / Zip or Gzip from Rare External Location

Anomalous Server Activity / Rare External from Server

Compromise / Agent Beacon (Long Period)

Compromise / Agent Beacon (Medium Period)

Compromise / Agent Beacon to New Endpoint

Compromise / Beacon for 4 Days

Compromise / Beacon to Young Endpoint

Compromise / Beaconing Activity To External Rare

Compromise / High Priority Tunnelling to Bin Services

Compromise / High Volume of Connections with Beacon Score

Compromise / HTTP Beaconing to New IP

Compromise / HTTP Beaconing to Rare Destination

Compromise / Large Number of Suspicious Failed Connections

Compromise / Large Number of Suspicious Successful Connections

Compromise / Slow Beaconing Activity To External Rare

Compromise / SSL Beaconing to Rare Destination

Compromise / Suspicious Beaconing Behavior

Compromise / Suspicious File and C2

Compromise / Suspicious HTTP and Anomalous Activity

Compromise / Suspicious TLS Beaconing To Rare External

Compromise / Sustained SSL or HTTP Increase

Compromise / Sustained TCP Beaconing Activity To Rare Endpoint

Device / Initial Attack Chain Activity

Device / New User Agent

MITRE ATT&CK Mapping

Tactic – Technique

INITIAL ACCESS – Exploit Public-Facing Application

RESOURCE DEVELOPMENT – Malware

EXECUTION – Scheduled Task/Job (Cron)

EXECUTION – Unix Shell

PERSISTENCE – Web Shell

DEFENSE EVASION – Masquerading (Masquerade File Type)

DEFENSE EVASION - Deobfuscate/Decode Files or Information

CREDENTIAL ACCESS – Brute Force

DISCOVERY – Remote System Discovery

COMMAND AND CONTROL – Ingress Tool Transfer

COMMAND AND CONTROL – Application Layer Protocol (Web Protocols)

COMMAND AND CONTROL – Encrypted Channel

COMMAND AND CONTROL – Non-Standard Port

COMMAND AND CONTROL – Data Obfuscation

IMPACT – Resource Hijacking (Compute)

List of IoCs

IoC         –          Type         –        Description

  • sys.traceroute[.]vip     – Hostname - C2 Endpoint
  • 77.221.158[.]154     – IP - C2 Endpoint
  • 185.174.137[.]26     – IP - C2 Endpoint
  • 93.113.25[.]46     – IP - C2 Endpoint
  • 104.131.69[.]106     – IP - C2 Endpoint
  • 95.164.5[.]41     – IP - C2 Endpoint
  • bristol-beacon-assets.s3.amazonaws[.]com     – Hostname - Payload Server
  • img.dxyjg[.]com     – Hostname - Payload Server
  • 38.180.147[.]18     – IP - Payload Server
  • 143.198.1[.]178     – IP - Payload Server
  • 185.208.156[.]46     – IP - Payload Server
  • 185.196.9[.]154     – IP - Payload Server
  • 46.8.226[.]75     – IP - Payload Server
  • 223.165.4[.]175     – IP - Payload Server
  • 188.166.244[.]81     – IP - Payload Server
  • bristol-beaconassets.s3[.]amazonaws[.]com/Y5bHaYxvd84sw     – URL - Payload
  • img[.]dxyjg[.]com/KjQfcPNzMrgV     – URL - Payload
  • 38.180.147[.]18/palofd     – URL - Payload
  • 90f6890fa94b25fbf4d5c49f1ea354a023e06510     – SHA1 - Associated to file /palofd
  • 143.198.1[.]178/7Z0THCJ     – URL - Payload
  • 8d82ccdb21425cf27b5feb47d9b7fb0c0454a9ca     – SHA1 - Associated to file /7Z0THCJ
  • fefd0f93dcd6215d9b8c80606327f5d3a8c89712     – SHA1 - Associated to file /7Z0THCJ
  • e5464f14556f6e1dd88b11d6b212999dd9aee1b1     – SHA1 - Associated to file /7Z0THCJ
  • 143.198.1[.]178/o4VWvQ5pxICPm     – URL - Payload
  • 185.208.156[.]46/lUuL095knXd62DdR6umDig     – URL - Payload
  • 185.196.9[.]154/ykKDzZ5o0AUSfkrzU5BY4w     – URL - Payload
  • 46.8.226[.]75/1.txt     – URL - Payload
  • 223.165.4[.]175/x6     – URL - Payload
  • 45.76.141[.]166/y.sh     – URL - Payload
  • repositorylinux[.]org/linux.sh     – URL - Payload
  • repositorylinux[.]org/cron.sh     – URL - Payload

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
Adam Potter
Senior Cyber Analyst

More in this series

No items found.

Blog

/

/

May 20, 2026

Prompt Security in Enterprise AI: Strengths, Weaknesses, and Common Approaches

prompt securityDefault blog imageDefault blog image

How enterprise AI Agents are changing the risk landscape  

Generative AI Agents are changing the way work gets done inside enterprises, and subsequently how security risks may emerge. Organizations have quickly realized that providing these agents with wider access to tooling, internal information, and granting permissions for the agent to perform autonomous actions can greatly increase the efficiency of employee workflows.

Early deployments of Generative AI systems led many organizations to scope individual components as self-contained applications: a chat interface, a model, and a prompt, with guardrails placed at the boundary. Research from Gartner has shown that while the volume and scope of Agentic AI deployments in enterprise environments is rapidly accelerating, many of the mechanisms required to manage risk, trust, and cost are still maturing.

The issue now resides on whether an agent can be influenced, misdirected, or manipulated in ways that leads to unsafe behavior across a broader system.

Why prompt security matters in enterprise AI

Prompt security matters in enterprise AI because prompts are the primary way users and systems interact with Agentic AI models, making them one of the earliest and most visible indicators of how these systems are being used and where risk may emerge.

For security teams, prompt monitoring is a logical starting point for understanding enterprise AI usage, providing insight into what types of questions are being asked and tasks are being given to AI Agents, how these systems are being guided, and whether interactions align with expected behavior. Complete prompt security takes this one step further, filtering out or blocking sensitive or dangerous content to prevent risks like prompt injection and data leakage.

However, visibility only at the prompt layer can create a false sense of security. Prompts show what was asked, but not always why it was asked, or what downstream actions were triggered by the agent across connected systems, data sources, or applications.

What prompt security reveals  

The primary function of prompt security is to minimize risks associated with generative and agentic AI use, but monitoring and analysis of prompts can also grant insight into use cases for particular agents and model. With comprehensive prompt security, security teams should be able to answer the following questions for each prompt:

  • What task was the user attempting to complete?
  • What data was included in the request, and was any of the data high-risk or confidential?
  • Was the interaction high-risk, potentially malicious, or in violation of company policy?
  • Was the prompt anomalous (in comparison to previous prompts sent to the agent / model)?

Improving visibility at this layer is a necessary first step, allowing organizations to establish a baseline for how AI systems are being used and where potential risks may exist.  

Prompt security alone does not provide a complete view of risk. Further data is needed to understand how the prompt is interpreted, how context is applied, what autonomous actions the agent takes (if any), or what downstream systems are affected. Understanding the outcome of a query is just as important for complete prompt security as understanding the input prompt itself – for example, a perfectly normal, low-risk prompt may inadvertently result in an agent taking a high-risk action.

Comprehensive AI security systems like Darktrace / SECURE AI can monitor and analyze both the prompt submitted to a Generative AI system, as well as the responses and chain-of-thought of the system, providing greater insight into the behavior of the system. Darktrace / SECURE AI builds on the core Darktrace methodology, learning the expected behaviors of your organization and identifying deviations from the expected pattern of life.

How organizations address prompt security today

As prompt-level visibility has become a focus, a range of approaches have emerged to make this activity more observable and controllable. Various monitoring and logging tools aim to capture prompt inputs to be analyzed after the fact.  

Input validation and filtering systems attempt to intervene earlier, inspecting prompts before they reach the model. These controls look for known jailbreak patterns, language indicative of adversarial attacks, or ambiguous instructions which could push the system off course.

Importantly, for a prompt security solution to be accurate and effective, prompts must be continually observed and governed, rather than treated as a point-in-time snapshot.  

Where prompt security breaks down in real environments

In more complex environments, especially those involving multiple agents or extensive tool use, AI security becomes harder to define and control.

Agent-to-Agent communications can be harder to monitor and trace as these happen without direct user interaction. Communication between agents can create routes for potential context leakage between agents, unintentional privilege escalation, or even data leakage from a higher privileged agent to a lower privileged one.

Risk is shaped not just by what is asked, but by the conditions in which that prompt operates and the actions an agent takes. Controls at the orchestration layer are starting to reflect this reality. Techniques such as context isolation, scoped memory, and role-based boundaries aim to limit how far a prompt’s influence can extend.  

Furthermore, Shadow AI usage can be difficult to monitor. AI systems that are deployed outside of formal governance structures and Generative AI systems hosted on unknown endpoints can fly under the radar and can go unseen by monitoring tools, leaving a critical opening where adversarial prompts may go undetected. Darktrace / SECURE AI features comprehensive detection of Shadow AI usage, helping organizations identify potential risk areas.

How prompt security fits in a broader AI risk model

Prompt security is an important starting point, but it is not a complete security strategy. As AI systems become more integrated into enterprise environments, the risks extend to what resources the system can access, how it interprets context, and what actions it is allowed to take across connected tools and workflows.

This creates a gap between visibility and control. Prompt security alone allows security teams to observe prompt activity but falls short of creating a clear understanding of how that activity translates into real-world impact across the organization.

Closing that gap requires a broader approach, one that connects signals across human and AI agent identities, SaaS, cloud, and endpoint environments. It means understanding not just how an AI system is being used, but how that usage interacts with the rest of the digital estate.

Prompt security, in that sense, is less of a standalone solution and more of an entry point into a larger problem: securing AI across the enterprise as a whole.

Explore how Darktrace / SECURE AI brings prompt security to enterprises

Darktrace brings more than a decade of AI expertise, built on an enterprise‑wide platform designed to operate in and understand the behaviors of the complex, ambiguous environments where today’s AI now lives. With Darktrace / SECURE AI, enterprises can safely adopt, manage, monitor, and build AI within their business.  

Learn about Darktrace / SECURE AI here.

Sign up today to stay informed about innovations across securing AI.

[related-resource]

Continue reading
About the author
Jamie Bali
Technical Author (AI) Developer

Blog

/

/

May 20, 2026

State of AI Cybersecurity 2026: 77% of security stacks include AI, but trust is lagging

Default blog imageDefault blog image

Findings in this blog are taken from Darktrace’s annual State of AI Cybersecurity Report 2026.

AI is a contributing member of nearly every modern cybersecurity team. As we discussed earlier in this blog series, rapid AI adoption is expanding the attack surface in ways that security professionals have never before experienced while also empowering attackers to operate at unprecedented speed and scale. It’s only logical that defenders are harnessing the power of AI to fight back.

After all, AI can help cybersecurity teams spot the subtle signs of novel threats before humans can, investigate events more quickly and thoroughly, and automate response. But although AI has been widely adopted, this technology is also frequently misunderstood, and occasionally viewed with suspicion.

For CISOs, the cybersecurity marketplace can be noisy. Making sense of competing vendors’ claims to distinguish the solutions that truly deliver on AI’s full potential from those that do not isn’t always easy. Without a nuanced understanding of the different types of AI used across the cybersecurity stack, it is difficult to make informed decisions about which vendors to work with or how to gain the most value from their solutions. Many security leaders are turning to Managed Security Service Providers (MSSPs) for guidance and support.

The right kinds of AI in the right places?

Back in 2024, when we first conducted this annual survey, more than a quarter of respondents were only vaguely familiar with generative AI or hadn’t heard of it at all. Today, GenAI plays a role in 77% of security stacks. This percentage marks a rapid increase in both awareness and adoption over a relatively short period of time.

According to security professionals, different types of AI are widely integrated into cybersecurity tooling:

  • 67% report that their organization’s security stack uses supervised machine learning
  • 67% report that theirs uses agentic AI
  • 58% report that theirs uses natural language processing (NLP)
  • 35% report that theirs uses unsupervised machine learning

But their responses suggest that organizations aren’t always using the most valuable types of AI for the most relevant use cases.

Despite all the recent attention AI has gotten, supervised machine learning isn’t new. Cybersecurity vendors have been experimenting with models trained on hand-labeled datasets for over a decade. These systems are fed large numbers of examples of malicious activity – for instance, strains of ransomware – and use these examples to generalize common indicators of maliciousness – such as the TTPs of multiple known ransomware strains – so that the models can identify similar attacks in the future. This approach is more effective than signature-based detection, since it isn’t tied to an individual byte sequence or file hash. However, supervised machine learning models can miss patterns or features outside the training data set. When adversarial behavior shifts, these systems can’t easily pivot.

Unsupervised machine learning, by contrast, can identify key patterns and trends in unlabeled data without human input. This enables it to classify information independently and detect anomalies without needing to be taught about past threats. Unsupervised learning can continuously learn about an environment and adapt in real time.

One key distinction between supervised and unsupervised machine learning is that supervised learning algorithms require periodic updating and re-training, whereas unsupervised machine learning trains itself while it works.

The question of trust

Even as AI moves into the mainstream, security professionals are eyeing it with a mix of enthusiasm and caution. Although 89% say they have good visibility into the reasoning behind AI-generated outputs, 74% are limiting AI’s ability to take autonomous action in their SOC until explainability improves. 86% do not allow AI to take even small remediation actions without human oversight.

This model, commonly known as “human in the loop,” is currently the norm across the industry. It seems like a best-of-both-worlds approach that allows teams to experience the benefits of AI-accelerated response without relinquishing control – or needing to trust an AI system.

Keeping humans somewhat in the loop is essential for getting the best out of AI. Analysts will always need to review alerts, make judgement calls, and set guardrails for AI's behavior. Their input helps AI models better understand what “normal” looks like, improving their accuracy over time.

However, relying on human confirmation has real costs – it delays response, increases the cognitive burden analysts must bear, and creates potential coverage gaps when security teams are overwhelmed or unavailable. The traditional model, in which humans monitor and act on every alert, is no longer workable at scale.

If organizations depend too heavily on in-the-loop humans, they risk recreating the very problem AI is meant to solve: backlogs of alerts waiting for analyst review. Removing the human from the loop can buy back valuable time, which analysts can then invest in building a proactive security posture. They can also focus more closely on the most critical incidents, where human attention is truly needed.

Allowing AI to operate autonomously requires trust in its decision-making. This trust can be built gradually over time, with autonomous operations expanding as trust grows. But it also requires knowledge and understanding of AI — what it is, how it works, and how best to deploy it at enterprise scale.

Looking for help in all the right places

To gain access to these capabilities in a way that’s efficient and scalable, growing numbers of security leaders are looking for outsourced support. In fact, 85% of security professionals prefer to obtain new SOC capabilities in the form of a managed service.

This makes sense: Managed Security Service Providers (MSSPs) can deliver deep, continuously available expertise without the cost and complexity of building an in-house team. Outsourcing also allows organizations to scale security coverage up or down as needs change, stay current with evolving threats and regulatory requirements, and leverage AI-native detection and response without needing to manage the AI tools themselves.

Preferences for MSSP-delivered security operations are particularly strong in the education, energy (87%), and healthcare sectors. This makes sense: all are high-value targets for threat actors, and all tend to have limited cybersecurity budgets, so the need for a partner who can deliver affordable access to expertise at scale is strong. Retailers also voiced a strong preference for MSSP-delivered services. These companies are tasked with managing large volumes of consumer personal and financial data, and with transforming an industry traditionally thought of as a late adopter to a vanguard of cyber defense. Technology companies, too, have a marked preference for SOC capabilities delivered by MSSPs. This may simply be because they understand the complexity of the threat landscape – and the advantages of specialized expertise — so well.

In order to help as many organizations as possible – from major enterprises to small and midmarket companies – benefit from enterprise-grade, AI-native security, Darktrace is making it easier for MSSPs to deliver its technology. The ActiveAI Security Portal introduces an alert dashboard designed to increase the speed and efficiency of alert triage, while a new AI-powered managed email security solution is giving MSSPs an edge in the never-ending fight against advanced phishing attacks – helping partners as well as organizations succeed on the frontlines of cyber defense.

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