ブログ
/
Network
/
December 22, 2021

9 Stages of Ransomware & How AI Responds

Discover the 9 stages of ransomware attacks and how AI responds at each stage. Learn how you can protect your business from cyber 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
Dan Fein
VP, Product
Default blog image
22
Dec 2021

Ransomware gets its name by commandeering and holding assets ransom, extorting their owner for money in exchange for discretion and full cooperation in returning exfiltrated data and providing decryption keys to allow business to resume.

Average ransom demands are skyrocketing, rising to $5.3 million in 2021, a 518% increase from the previous year. But the cost of recovering from a ransomware attack typically far exceeds the ransom payments: the average downtime after a ransomware attack is 21 days; and 66% of ransomware victims report a significant loss of revenue following a successful attack.

In this series, we break down this huge topic step by step. Ransomware is a multi-stage problem, requiring a multi-stage solution that autonomously and effectively contains the attack at any stage. Read on to discover how Self-Learning AI and Autonomous Response stops ransomware in its tracks.

1. Initial intrusion (email)

Initial entry – the first stage of a ransomware attack – can be achieved through RDP brute-forcing (exposed Internet service), malicious websites and drive-by downloads, an insider threat with company credentials, system and software vulnerabilities, or any number of other attack vectors.

But the most common initial attack vector is email. An organization’s biggest security weakness is often their people – and attackers are good at finding ways of exploiting this. Well-researched, targeted, legitimate-looking emails are aimed at employees attempting to solicit a reaction: a click of a link, an opening of an attachment, or persuading them to divulge credentials or other sensitive information.

Gateways: Stops what has been seen before

Most conventional email tools rely on past indicators of attack to try and spot the next threat. If an email comes in from a blocklisted IP address or email domain, and uses known malware that has previously been seen in the wild, the attack may be blocked.

But the reality is, attackers know the majority of defenses take this historical approach, and so constantly update their attack infrastructure to bypass these tools. By buying new domains for a few pennies each, or creating bespoke malware with just small adaptions to the code, they can outpace and outsmart the legacy approach taken by a typical email gateway.

Real-world example: Supply chain phishing attack

By contrast, Darktrace’s evolving understanding of ‘normal’ for every email user in the organization enables it to detect subtle deviations that point to a threat – even if the sender or any malicious contents of the email are unknown to threat intelligence. This is what enabled the technology to stop an attack that recently targeted McLaren Racing, with emails sent to a dozen employees in the organization each containing a malicious link. This possible precursor to ransomware bypassed conventional email tools – largely because it was sent from a known supplier – however Darktrace recognized the account hijack and held the email back.

Figure 1: A snapshot of Darktrace’s Threat Visualizer surfacing the malicious email

Read the full case study

2. Initial intrusion (server-side)

With organizations rapidly expanding their Internet-facing perimeter, this increased attack surface has paved the way for a surge in brute-force and server-side attacks.

A number of vulnerabilities against such Internet-facing servers and systems have been disclosed this year, and for attackers, targeting and exploiting public-facing infrastructure is easier than ever – scanning the Internet for vulnerable systems is made simple with tools like Shodan or MassScan.

Attackers may also achieve initial intrusion via RDP brute-forcing or stolen credentials, with attackers often reusing legitimate credentials from previous data dumps. This has much higher precision and is less noisy than a classic brute-force attack.

A lot of ransomware attacks use RDP as an entry vector. This is part of a wider trend of ‘Living off the Land’: using legitimate off-the-shelf tools (abusing RDP, SMB1 protocol, or various command line tools WMI or Powershell) to blur detection and attribution by blending in with typical administrator activity. Ensuring that backups are isolated, configurations are hardened, and systems are patched is not enough – real-time detection of every anomalous action is needed.

Antivirus, firewalls and SIEMs

In cases of malware downloads, endpoint antivirus will detect these if, and only if, the malware has been seen and fingerprinted before. Firewalls typically require configuration on a per-organization basis, and often need to be modified based on the needs of the business. If the attack hits the firewall where a rule or signature does not match it, again, it will bypass the firewall.

SIEM and SOAR tools also look for known malware being downloaded, leverage pre-programmed rules and use pre-programmed responses. While these tools do look for patterns, these patterns are defined in advance, and this approach relies on a new attack to have sufficiently similar traits to attacks that have been seen before.

Real-world example: Dharma ransomware

Darktrace detected a targeted Dharma ransomware attack against a UK organization exploiting an open RDP connection through Internet-facing servers. The RDP server began receiving a large number of incoming connections from rare IP addresses on the Internet. It is highly likely that the RDP credential used in this attack had been compromised at a previous stage – either via common brute-force methods, credential stuffing attacks, or phishing. Indeed, a technique growing in popularity is to buy RDP credentials on marketplaces and skip to initial access.

Figure 2: The model breaches that fired over the course of this attack, including anomalous RDP activity

Unfortunately, in this case, without Autonomous Response installed, the Dharma ransomware attack continued until its final stages, where the security team were forced into the heavy-handed and disruptive action of pulling the plug on the RDP server midway through encryption.

Read the full case study

3. Establish foothold and C2

Whether through a successful phish, a brute-force attack, or some other method, the attacker is in. Now, they make contact with the breached device(s) and establish a foothold.

This stage allows attackers to control subsequent stages of the attack remotely. During these command and control (C2) communications, further malware may also pass from the attacker to the devices. This helps them to establish an even greater foothold within the organization and readies them for lateral movement.

Attackers can adapt malware functionality with an assortment of ready-made plug-ins, allowing them to lie low inside the business undetected. More modern and sophisticated ransomware is able to adapt by itself to the surrounding environment, and operate autonomously, blending in to regular activity even when cut off from its command and control server. These ‘self-sufficient’ ransomware strains pose a big problem for traditional defenses reliant on stopping threats solely on the grounds of its malicious external connections.

Viewing connections in isolation vs understanding the business

Conventional security tools like IDS and firewalls tend to look at connections in isolation rather than in the context of previous and potentially relevant connections, making command and control very difficult to spot.

IDS and firewalls may block ‘known-bad’ domains or use some geo-blocking, but this is where an attacker would likely leverage new infrastructure.

These tools also don’t tend to analyze for things like the periodicity, such as whether a connection is beaconing at a regular or irregular interval, or the age and rarity of the domain in the context of the environment.

With Darktrace’s evolving understanding of the digital enterprise, suspicious C2 connections and the downloads which follow them are spotted, even when conducted using regular programs or methods. The AI technology correlates multiple subtle signs of threat – a small subset of which includes anomalous connections to young and/or unusual endpoints, anomalous file downloads, incoming remote desktop, and unusual data uploads and downloads.

Once they are detected as a threat, Darktrace's Autonomous Response halts these connections and downloads, while allowing normal business activity to continue.

Real-world example: WastedLocker attack

When a WastedLocker ransomware attack hit a US agricultural organization, Darktrace immediately detected the initial unusual SSL C2 activity (based on a combination of destination rarity, JA3 unusualness and frequency analysis). Antigena (on this occasion configured in passive mode, and therefore not granted permission to take autonomous action) suggested instantly blocking the C2 traffic on port 443 and parallel internal scanning on port 135.

Figure 3: The Threat Visualizer reveals the action Antigena would have taken

When beaconing was later observed to bywce.payment.refinedwebs[.]com, this time over HTTP to /updateSoftwareVersion, Antigena escalated its response by blocking the further C2 channels.

Figure 4: Antigena escalates its response

Read the full case study

4. Lateral movement

Once an attacker has established a foothold within an organization, they begin to increase their knowledge of the wider digital estate and their presence within it. This is how they will find and access the files which they will ultimately attempt to exfiltrate and encrypt. It begins reconnaissance: scanning the network; building up a picture of its component devices; identifying the location of the most valuable assets.

Then the attacker begins moving laterally. They infect more devices and look to escalate their privileges – for instance, by obtaining admin credentials – thereby increasing their control over the environment. Once they have obtained authority and presence within the digital estate, they can progress to the final stages of the attack.

Modern ransomware has built-in functions that allow it to search automatically for stored passwords and spread through the network. More sophisticated strains are designed to build themselves differently in different environments, so the signature is constantly changing and it’s harder to detect.

Legacy tools: A blunt response to known threats

Because they rely upon static rules and signatures, legacy solutions struggle to prevent lateral movement and privilege escalation without also impeding essential business operations. Whilst in theory, an organization leveraging firewalls and NAC internally with proper network segmentation and a perfect configuration could prevent cross-network lateral movement, maintaining a perfect balance between protective and disruptive controls is near impossible.

Some organizations rely on Intrusion Prevent Systems (IPS) to deny network traffic when known threats are detected in packets, but as with previous stages, novel malware will evade detection, and this requires the database to be constantly updated. These solutions also sit at the ingress/egress points, limiting their network visibility. An Intrusion Detection System (IDS) may sit out-of-line, but doesn’t have response capabilities.

A self-learning approach

Darktrace’s AI learns ‘self’ for the organization, enabling it to detect suspicious activity indicative of lateral movement, regardless of whether the attacker uses new infrastructure or ‘lives off the land’. Potential unusual activity that Darktrace detects includes unusual scanning activity, unusual SMB, RDP, and SSH activity. Other models that fire at this stage include:

  • Suspicious Activity on High-Risk Device
  • Numeric EXE in SMB Write
  • New or Uncommon Service Control

Autonomous Response then takes targeted action to stop the threat at this stage, blocking anomalous connections, enforcing the infected device’s ‘pattern of life’, or enforcing the group ‘pattern of life’ – automatically clustering devices into peer groups and preventing a device from doing anything its peer group hasn’t done.

Where malicious behavior persists, and only if necessary, Darktrace will quarantine an infected device.

Real-world example: Unusual chain of RDP connections

At an organization in Singapore, one compromised server led to the creation of a botnet, which began moving laterally, predominantly by establishing chains of unusual RDP connections. The server then started making external SMB and RPC connections to rare endpoints on the Internet, in an attempt to find further vulnerable hosts.

Other lateral movement activities detected by Darktrace included the repeated failing attempts to access multiple internal devices over the SMB file-sharing protocol with a range of different usernames, implying brute-force network access attempts.

Figure 5: Darktrace’s Cyber AI Analyst reveals suspicious TCP scanning followed by a suspicious chain of administrative RDP connections

Read the full case study

5. Data exfiltration

In the past, ransomware was simply about encrypting an operating system and network files.

In a modern attack, as organizations insure against malicious encryption by becoming increasingly diligent with data backups, threat actors have moved towards ‘double extortion’, where they exfiltrate key data and destroy backups before the encryption takes place. Exfiltrated data is used to blackmail organizations, with attackers threatening to publish sensitive information online or sell it on to the organization’s competitors if they are not paid.

Modern ransomware variants also look for cloud file storage repositories such as Box, Dropbox, and others.

Many of these incidents aren’t public, because if IP is stolen, organizations are not always legally required to disclose it. However, in the case of customer data, organizations are obligated by law to disclose the incident and face the additional burden of compliance files – and we’ve seen these mount in recent years (Marriot, $23.8 million; British Airways, $26 million; Equifax, $575 million). There’s also the reputational blow associated with having to inform customers that a data breach has occurred.

Legacy tools: The same old story

For those that have been following, the narrative by now will sound familiar: to stop a ransomware attack at this stage, most defenses rely on either pre-programmed definitions of 'bad' or have rules constructed to combat different scenarios put organizations in a risky, never-ending game of cat and mouse.

A firewall and proxy might block connections based on pre-programmed policies based on specific endpoints or data volumes, but it’s likely an attacker will ‘live off the land’ and utilize a service that is generally allowed by the business.

The effectiveness of these tools will vary according to data volumes: they might be effective for ‘smash and grab’ attacks using known malware, and without employing any defense evasion techniques, but are unlikely to spot ‘low and slow’ exfiltration and novel or sophisticated strains.

On the other hand, because by nature it involves a break from expected behavior, even less conspicuous, low and slow data exfiltration is detected by Darktrace and stopped with Darktrace's Autonomos Response. No confidential files are lost, and attackers are unable to extort a ransom payment through blackmail.

Real-world example: Unusual chain of RDP connections

It becomes more difficult to find examples of Darktrace stopping ransomware at these later stages, as the threat is usually contained before it gets this far. This is the double-edged sword of effective security – early containment makes for bad storytelling! However, we can see the effects of a double extortion ransomware attack on an energy company in Canada. The organization had the Enterprise Immune System but no Antigena, and without anyone actively monitoring Darktrace’s AI detections, the attack was allowed to unfold.

The attacker managed to connect to an internal file server and download 1.95TB of data. The device was also seen downloading Rclone software – an open-source tool, which was likely applied to sync data automatically to the legitimate file storage service pCloud. Following the completion of the data exfiltration, the device ‘serverps’ finally began encrypting files on 12 devices with the extension *.06d79000. As with the majority of ransomware incidents, the encryption happened outside of office hours – overnight in local time – to minimize the chance of the security team responding quickly.

Read the full details of the attack

It should be noted that the exact order of the stages 3–5 above is not set in stone, and varies according to attack. Sometimes data is exfiltrated and then there is further lateral movement, and additional C2 beaconing. This entire period is known as the ‘dwell time’. Sometimes it takes place over only a few days, other times attackers may persist for months, slowly gathering more intel and exfiltrating data in a ‘low and slow’ fashion so as to avoid detection from rule-based tools that are configured to flag any single data transfer over a certain threshold. Only through a holistic understanding of malicious activity over time can a technology spot this level of activity and allow the security team to remove the threat before it reaches the latter and most damaging stages of ransomware.

6. Data encryption

Using either symmetric encryption, asymmetric encryption, or a combination of the two, attackers attempt to render as much data unusable in the organization’s network as they can before the attack is detected.

As the attackers alone have access to the relevant decryption keys, they are now in total control of what happens to the organization’s data.

Pre-programmed response and disruption

There are many families of tools that claim to stop encryption in this manner, but each contain blind spots which enable a sophisticated attacker to evade detection at this crucial stage. Where they do take action, it is often highly disruptive, causing major shutdowns and preventing a business from continuing its usual operations.

Internal firewalls prevent clients from accessing servers, so once an attacker has penetrated to servers using any of the techniques outlined above, they have complete freedom to act as they want.

Similarly, antivirus tools look only for known malware. If the malware has not been detected until this point, it is highly unlikely the antivirus will act here.

Stopping encryption autonomously

Even if familiar tools and methods are used to conduct it, Autonomous Response can enforce the normal ‘pattern of life’ for devices attempting encryption, without using static rules or signatures. This action can be taken independently or via integrations with native security controls, maximizing the return on other security investments. With a targeted Autonomous Response, normal business operations can continue while encryption is prevented.

7. Ransom note

It is important to note that in the stages before encryption, this ransomware attack is not yet “ransomware”. Only at this stage does it gets its name.

A ransom note is deployed. The attackers request payment in return for a decryption key and threaten the release of sensitive exfiltrated data. The organization must decide whether to pay the ransom or lose their data, possibly to their competition or the public. The average demand made by ransomware threat actors rose in 2021 to $5.3 million, with meat processing company JBS paying out $11 million and DarkSide receiving over $90 million in Bitcoin payments following the Colonial Pipeline incident.

All of the stages up until this point represent a typical, traditional ransomware attack. But ransomware is shifting from indiscriminate encryption of devices to attackers targeting business disruption in general, using multiple techniques to hold their victims to ransom. Additional methods of extortion include not only data exfiltration, but corporate domain hijack, deletion or encryption of backups, attacks against systems close to industrial control systems, targeting company VIPs… the list goes on.

Sometimes, attackers will just skip straight from stage 2 to 6 and jump straight to extortion. Darktrace recently stopped an email attack which showed an attacker bypassing the hard work and attempting to jump straight to extortion in an email. The attacker claimed to have compromised the organization’s sensitive data, requesting payment in bitcoin for its same return. Whether or not the claims were true, this attack shows that encryption is not always necessary for extortion, and this type of harassment exists in multiple forms.

Figure 6: Darktrace holds back the offending email, protecting the recipient and organization from harm

As with the email example we explored in the first post of this series, Darktrace/Email was able to step in and stop this email where other email tools would have let it through, stopping this potentially costly extortion attempt.

Whether through encryption or some other kind of blackmail, the message is the same every time. Pay up, or else. At this stage, it’s too late to start thinking about any of the options described above that were available to the organization, that would have stopped the attack in its earliest stages. There is only one dilemma. “To pay or not to pay” – that is the question.

Often, people believe their payment troubles are over after the ransom payment stage, but unfortunately, it’s just beginning to scratch the surface…

8. Clean-up

Efforts are made to try to secure the vulnerabilities which allowed the attack to happen initially – the organization should be conscious that approximately 80% of ransomware victims will in fact be targeted again in the future.

Legacy tools largely fail to shed light on the vulnerabilities which allowed the initial breach. Like searching for a needle in an incomplete haystack, security teams will struggle to find useful information within the limited logs offered by firewalls and IDSs. Antivirus solutions may reveal some known malware but fail to spot novel attack vectors.

With Darktrace’s Cyber AI Analyst, organizations are given full visibility over every stage of the attack, across all coverage areas of their digital estate, taking the mystery out of ransomware attacks. They are also able to see the actions that would have been taken to halt the attack by Darktrace RESPOND.

9. Recovery

The organization begins attempts to return its digital environment to order. Even if it has paid for a decryption key, many files may remain encrypted or corrupted. Beyond the costs of the ransom payment, network shutdowns, business disruption, remediation efforts, and PR setbacks all incur hefty financial losses.

The victim organization may also suffer additional reputation costs, with 66% of victims reporting a significant loss of revenue following a ransomware attack, and 32% reporting losing C-level talent as a direct result from ransomware.

Conclusion

While the high-level stages described above are common in most ransomware attacks, the minute you start looking at the details, you realize every ransomware attack is different.

As many targeted ransomware attacks come through ransomware affiliates, the Tools, Techniques and Procedures (TTPs) displayed during intrusions vary widely, even when the same ransomware malware is used. This means that even comparing two different ransomware attacks using the same ransomware family, you are likely to encounter completely different TTPs. This makes it impossible to predict what tomorrow’s ransomware will look like.

This is the nail in the coffin for traditional tooling which is based on historic attack data. The above examples demonstrate that Self-Learning technology and Autonomous Response is the only solution that stops ransomware at every stage, across email and network.

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

More in this series

No items found.

Blog

/

Network

/

June 24, 2026

From Click to Command: Behavioral Detection of AppleScript-Led MacOS Intrusions

Default blog imageDefault blog image

Introduction

Darktrace’s Threat Research team is publishing this analysis to help defenders understand an active pattern of macOS tradecraft observed in multiple customer environments. This post summarizes the behaviors observed, how they were assessed, and what defenders can do now.

Across multiple environments, Darktrace observed a consistent MacOS intrusion pattern beginning with ClickFix-style user-assisted “update” execution and transitioning into AppleScript-driven post-compromise activity and sustained outbound signaling.

While individual indicators were low-confidence, the repeated convergence of weak behavioral signals — including HTTP POST beaconing, rare or IP-only destinations, SSL anomalies, and abnormal client characteristics — provided a defensible indication of command-and-control establishment Darktrace detection and response in these cases was driven by behavior over artifacts. In the highest-confidence instances, automated containment disrupted outbound signaling before sustained tasking could occur.

Background

ClickFix-style activity typically relies on user-assisted execution and plausible “update” pretexting, followed by post-execution use of native tools to keep the footprint light. In MacOS environments, AppleScript and other built-in scripting mechanisms enable flexible post-compromise workflows while minimizing stable file-based indicators.

Following execution, affected devices exhibited a consistent behavioral pattern. AppleScript or equivalent native scripting activity was observed initiating follow-on workflows, after which outbound communications began to establish a structured rhythm.

These communications were characterized by repeated HTTP POST requests to low-prevalence or IP-only endpoints, often combined with unusual SSL properties and client identifiers that diverged from baseline device behavior. Individually, these signals were weak. When correlated across time and devices, they formed a pattern consistent with control establishment rather than benign software activity.

In higher-confidence cases, Autonomous Response actions were able to reduce or halt outbound signaling, interrupting the attacker’s ability to maintain control.

Detection Timeline

In representative cases, the sequence unfolded as follows:

Stage 1 – Initial Execution

Initial activity began with suspicious or masqueraded execution on a MacOS endpoint, consistent with ClickFix-style user deception.

Stage 2 – Post-Execution Scripting

This was followed closely by native scripting activity, most commonly AppleScript, indicating the transition into post-execution workflow.

Stage 3 – Outbound Communications

Outbound communications then emerged, initially sporadic but quickly forming a consistent cadence of HTTP POST requests to rare external endpoints.

Stage 4 – Anomaly Convergence

As activity persisted, additional anomalies became visible — unusual SSL characteristics, abnormal user agents, and connections to infrastructure with no prior network prevalence.

Stage 5 – Autonomous Response

In the most mature stages of the activity, automated containment actions disrupted outbound communications on affected devices, limiting the attacker’s ability to continue tasking while investigations progressed.

Darktrace coverage and detections

The following use-case highlights systems likely affected by malicious macOS intrusion activity linked by Microsoft to the Democratic People’s Republic of Korea (DPRK) [1], with indications of suspicious behavior observed between March 1 and May 3, 2026. The activity overlaps with patterns described in recent reporting on DPRK-nexus MacOS intrusions [1], though attribution confidence in this case remains moderate and based on behavioral alignment rather than solely infrastructure linkage.

Analyst confidence emerged through the correlation of multiple weak signals across time and devices. This included model coverage for rare external communications, sustained beaconing patterns, repeated HTTP POSTs, and anomalous client characteristics. Where enabled, Autonomous Response actions disrupted the most active outbound paths to reduce the attacker’s ability to maintain control while Darktrace’s investigation continued.

Notably, this highly anomalous behavior included:

  • Outbound connections to the rare external endpoint, zoom[.]uswebob[.]us associated with IP address, 148.72.73[.]98 [2][3] over port 443
  • Outbound connections to the rare external endpoint, check02id[.]com associated with IP address, 83.136.210[.]180 [4] over port 7365
  • Outbound connections to the rare external endpoints, 104.145.210[.]107 [5] over port 8443 and 83.136.208[.]48 [6] over port 443
  • Outbound connections to the rare external endpoint, 83.136.208[.]246 [7] over port 6783 with observed URI `/api/daemon` and a PowerShell user agent

Darktrace’s detection initially highlighted a desktop device (running MacOS) engaging in anomalous behavior as early as March 12, 2026. Starting on March 12, the source device triggered a ‘Possible Doppelganger Attack’ alert including connectivity to the hostname "zoom[.]uswebob[.]us · 148.72.73[.]98" over port 443 (TCP, HTTPS, H2). This model highlights a device connecting to a location that is rare but masquerades as legitimate software, such as Zoom in this case, a commonly used technique to blend into expected traffic [2] [3].

 Initial connectivity observed to the rare external hostname, zoom[.]uswebob[.]us · 148.72.73[.]98, over port 443.
Figure 1: Initial connectivity observed to the rare external hostname, zoom[.]uswebob[.]us · 148.72.73[.]98, over port 443.

This was followed roughly seven later by a connection to 104.145.210[.]107 over port 8443, during which approximately 250 KiB of data of inbound data and 30 MiB of outbound data was observed, triggering the ‘Unusual Activity / Unusual External Data to New Endpoint’ in Darktrace.

Quickly after this connection, Darktrace’s Autonomous Response intervened, blocking the device’s access to the unusual external location and halting the data exfiltration attempt.

Figure 2: Darktrace’s detection of unusual data exfiltration, shortly followed by an Autonomous Response action to block it.

The device continued to consistently trigger model alerts relating to unusual external connectivity, including 'Posting HTTP to IP Without Hostname', 'Anomalous Connection / Rare External SSL Self-Signed' alerts, until well after 3 PM that day.

Figure 3: Additional external connectivity to new IP without a hostname, including connectivity to 83.136.208[.]246, alongside an anomalous ‘curl/8.7.1’ user agent and ‘/api/daemon’ URI.
Figure 4: Continued external SSL connectivity to IP 83.136.208[.]48, including connectivity to 83.136.208[.]246, alongside an anomalous ‘curl/8.7.1’ user agent and ‘/api/daemon’ URI.
Figure 5: Continued external HTTP connectivity to hostname, check02id[.]com · 83.136.210[.]180, alongside an anomalous ‘Go-http-client/1,1’ user agent.

From March 13 to March 28, the device continued exhibit unusual connectivity to various endpoints (e.g., 83.136.208[.]48, 83.136.208[.]246, check02id[.]com · 83.136.210[.]180), with the 'Multiple HTTP POSTs to Rare Hostname' model consistently triggering.

Windows OS Case

Pivoting over to an additional device, this time running Windows OS, anomalous behavior was also observed between March 30 and April 20. Notably, on March 30, the device was observed making a large number of suspicious external connection attempts to 83.136.208[.]246 over port 6783, all of which failed.

A further indicator was observed on April 1 with PowerShell connectivity to the same rare endpoint (83.136.208[.]246, port 6783), using the URI '/api/daemon' and the user agent 'Mozilla/5.0 (Windows NT; Windows NT 10.0; fr-FR) WindowsPowerShell/5.1.26100.7920'.  Additional alerts included 'New User Agent to IP Without Hostname' and 'Anomalous Github Download', alongside activity involving the same endpoint.

Figure 6 : ‘Anomalous Powershell to Rare External Destination’ and ‘Github Download’ model alerts. This behavior involved connectivity with the endpoints ‘83.136.208[.]246’ and ‘github[.]com’.

The device continued triggering 'Posting HTTP to IP Without Hostname' & 'PowerShell to External Rare' alerts between April 4 and April 20 across multiple related endpoints (i.e., 83.136.208[.]48, 83.136.208[.]246, check02id[.]com · 83.136.210[.]180).

Darktrace’s Autonomous Response capability was able to block suspicious PowerShell attempts to unusual external locations, as shown below in an example from April 20.

Figure 7:  Autonomous Response intervening to block an unusual PowerShell connection to an external destination.

Cyber AI Analyst investigations

In higher-confidence instances, Darktrace’s Cyber AI Analyst investigations helped connect otherwise separate model alerts into a single incident narrative, highlighting the attacker’s progression from post-execution scripting into sustained outbound signaling. This contextual stitching is particularly valuable in macOS scenarios where static artefacts are limited, and behavioral sequencing defines the intrusion.

Cyber AI Analyst investigations highlighted alerts on March 12, including unusual repeated connections and possible SSL command-and-control (C2) to multiple endpoints:

Figure 8: Cyber AI Analyst investigation linking events into a unified incident.

Autonomous Response

In addition to the containment actions detailed earlier, Autonomous Response implemented multiple additional measures to contain suspicious activity throughout the course of this attack. Whenever unusual external connectivity was detected, Darktrace blocked it, closing down potential C2 channels. Likewise, when data exfiltration attempts were identified, these connections were stopped to prevent the potential loss of sensitive data.

Figure 9: Autonomous Response actions implemented by Darktrace in response to suspicious connectivity in mid-March.

Furthermore, in cases where a device was deemed to have carried out a significant number of anomalous activities, Darktrace enforced a “pattern of life” on the device, preventing it from deviating from its expected behavior while allowing legitimate business operations to continue uninterrupted.

Figure 10: Autonomous Response actions implemented by Darktrace in response to suspicious connectivity in April, including the “Enforce Pattern of Life” action.

Conclusion

macOS intrusion tradecraft continues to shift toward native tooling and lightweight control channels designed to evade signature-led controls.

The repeated convergence of rare destinations, POST-based signaling, and anomalous client behavior — observed across time and across devices — provided sufficient evidence to act early and with confidence.

As macOS tradecraft continues to evolve, the defender advantage increasingly lies not in signatures, but in the ability to reason from behavior.

Credit to Justin Torres (Senior Cyber Analyst), Nathaniel Jones (VP, Security & AI Strategy, FCISO)

Edited by Ryan Traill (Content Manager)

Appendices

Darktrace Model Alert Coverage:

/ NETWORK-based model alerts:

·       Anomalous Connection::Multiple HTTP POSTs to Rare Hostname

·       Anomalous Connection::Rare External SSL Self-Signed

·       Anomalous Connection::Powershell to Rare External

·       Anomalous Connection::New User Agent to IP Without Hostname

·       Anomalous Connection::Posting HTTP to IP Without Hostname

·       Compromise::Fast Beaconing to DGA

·       Compromise::Large Number of Suspicious Failed Connections

·       Device::Anomalous Github Download

·       Device::New PowerShell User Agent

·       Unusual Activity::Unusual External Data to New Endpoint

/ NETWORK-based Autonomous Response model alerts:

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

·       Antigena / Network::Significant Anomaly::Antigena Controlled and Model Breach

·       Antigena / Network::Significant Anomaly::Antigena Breaches Over Time Block

Indicators of Compromise (IoCs)

IP/Hostname:

·       zoom[.]uswebob[.]us · 148.72.73[.]98

·       83.136.208[.]246

·       check02id[.]com · 83.136.210[.]180

·       83.136.208[.]48

·       104.145.210[.]107

URIs:

·       /api/daemon

Destination Port Usage:

·       6783

·       5202

·       443

·       7365

·       8443

ASN:

·       AS400897 PETROSKY

·       AS398256 AS-ULTAHOST

User agents:

·       Mozilla/5.0 (Windows NT; Windows NT 10.0; fr-FR) WindowsPowerShell/5.1.26100.7920

·       Go-http-client/1.1

·       curl/8.7.1

MITRE ATT&CK Mapping

(Technique Name - Tactic - ID - Sub-Technique of)

·       Browser Session Hijacking - COLLECTION - T1185

·       Web Protocols - COMMAND AND CONTROL - T1071.001 - T1071

·       Install Digital Certificate - RESOURCE DEVELOPMENT - T1608.003 - T1608

·       PowerShell - EXECUTION - T1059.001 - T1059

·       Domain Generation Algorithms - COMMAND AND CONTROL - T1568.002 - T1568

·       Non-Standard Port - COMMAND AND CONTROL - T1571

·       Malware - RESOURCE DEVELOPMENT - T1588.001 - T1588

·       Web Service - COMMAND AND CONTROL - T1102

·       Code Repositories - COLLECTION - T1213.003 - T1213

·       Exploitation of Remote Services - LATERAL MOVEMENT - T1210

·       Exfiltration Over C2 Channel - EXFILTRATION - T1041

·       Exfiltration to Cloud Storage - EXFILTRATION - T1567.002 - T1567

References:

[1] https://www.microsoft.com/en-us/security/blog/2026/04/16/dissecting-sapphire-sleets-macos-intrusion-from-lure-to-compromise/

[2] https://radar.securityalliance.org/advisory-on-dprk-unc1069-fake-microsoft-teams-and-zoom-calls/

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

[4] https://www.virustotal.com/gui/ip-address/83.136.210.180/community

[5] https://www.virustotal.com/gui/ip-address/104.145.210.107/community

[6] https://www.virustotal.com/gui/ip-address/83.136.208.48/community

[7] https://www.virustotal.com/gui/ip-address/83.136.208.246/community

[8] https://www.darktrace.com/blog/applescript-abuse-unpacking-a-macos-phishing-campaign

Continue reading
About the author
Justin Torres
Cyber Analyst

Blog

/

AI

/

June 24, 2026

A New Security Challenge: The Curious Case of Prompt Language Analysis

Default blog imageDefault blog image

Why prompt analysis is emerging as a key AI security challenge

If securing AI has been one of the defining cybersecurity conversations of the past year, prompt analysis is quickly becoming one of its most interesting frontiers.

Security leaders are under pressure to understand how AI is being used across the business. In some organizations, that means governing employee use of chatbots. In others, it means overseeing copilots embedded into SaaS platforms, monitoring coding assistants, or assessing the growing footprint of autonomous agents. However different these use cases may appear on the surface, they share a common factor: humans and machines are usually interacting with enterprise systems through language.  

How prompt language differs from traditional security telemetry

For years, defenders have become used to working with familiar forms of telemetry: email traffic, network connections, API calls, endpoint processes, authentication events. Prompt language is different. It is not simply another log source. It is an expression of intent, instruction, curiosity, urgency, and sometimes manipulation. It reflects the end-goal of a user or agent, but not always with enough surrounding context to interpret the risk correctly.

Why existing security approaches only partially explain prompt risk

A growing number of vendors are approaching the task of securing AI from the angle they know best. Perimeter vendors are extending web or browser controls into AI usage. Identity vendors are emphasizing agent permissions and access governance. Data security and DLP providers are focusing on content inspection and exfiltration risk. All of these perspectives matter, but individually can’t fully explain the problem.

The challenge with securing AI is not just that a new application category has emerged. It is that language has become a new operating layer in the enterprise.

Employees now use prompts to summarize documents, generate code, analyze spreadsheets, query internal knowledge, and trigger multi-step actions through agents. In each case, prompt language acts as the interface between human intent and machine execution. That makes prompts incredibly valuable from a security perspective as they can hint at misuse, policy violations, data exposure, or attempts to circumvent controls. However, they can also be deeply ambiguous when viewed in isolation. That ambiguity is the heart of the issue.

Prompts as behavioral signals, not just text to classify

A prompt by itself tells you what was asked. It does not necessarily tell you whether the request is expected, risky, accidental, or entirely legitimate in context. Two nearly identical prompts can carry very different meanings depending on the role and function of who issued them, what systems they can access, and what actions followed. In other words, prompts are not just text to classify. They are behavioral signals to interpret.

Example: How context changes prompt risk entirely

Consider a common enterprise scenario. An employee is pulled into a new project with an aggressive deadline. Almost overnight, their use of AI tools spikes. They begin prompting more frequently, working across unfamiliar documents, querying new data sources, and interacting with more systems than usual to accelerate delivery. Viewed narrowly, this may look suspicious. Prompt volume increases, file access patterns change, API and SaaS activity rise. From some vantage points, it may resemble insider risk or unmanaged AI usage.

But now add context. Imagine that, earlier that day, the employee received instructions from a senior leader asking them to support a time-sensitive initiative. Their communication history shows that this leader is a legitimate reporting-line superior. Their recent collaboration patterns align with the new project team. Their subsequent activity, while unusual for that individual’s baseline, is consistent with the business task they were assigned.

What initially looked like a risk event may actually be a normal response to business pressure. Without the surrounding context of communication, organizational relationships, and broader behavioral patterns, prompt activity alone could generate more noise than insight.

The reverse is also true. A prompt may appear benign on the surface while the context around it suggests elevated risk. A request that seems routine could originate from a compromised user, a newly connected external agent, a shadow AI workflow, or a user acting outside their normal role. The language itself may not contain anything obviously malicious, but the surrounding conditions may tell a very different story.

What security teams need to analyze prompts effectively

The future of prompt analysis is not just about understanding language. It is about understanding language in context.

To do that well, security teams need more than prompt inspection. They need to understand:

  • Who is issuing the prompt, whether human or agent
  • How that identity normally behaves across the enterprise
  • What systems, data, and workflows are connected to the interaction
  • Which relationships and communications explain the surrounding activity
  • Whether the downstream actions align with expected business behavior

When those layers are absent, prompt analysis can become another isolated control surface: useful in theory, but limited in practice. Security teams may detect unusual wording but miss the operational function behind it, overreact to benign changes in behavior, or miss subtle misuse because the prompt itself did not appear dangerous.

How organizations should think about prompt analysis going forward

Security teams have seen this pattern before. In the cloud, posture without runtime context left important gaps. In identity, access control without behavioral understanding missed misuse that looked legitimate on paper. In data security, content inspection without business context often created friction without resolving risk. AI is exposing the same lesson again: controls are strongest when they are coordinated, not isolated. As organizations work to secure AI and identify gaps across their security operations, prompt analysis will become an increasingly important source of insight, but only as part of a broader strategy.

Prompt analysis will undoubtedly become more common, as prompts are one of the clearest windows into how people and agents are using AI systems. However, what matters most is not simply collecting prompts or filtering dangerous phrases, but being able to place that language inside a wider behavioral and operational picture.

Organizations that already have a broader understanding of how work gets done across the enterprise will be better positioned to make sense of prompt language as this category matures. They will be better able to distinguish urgency from abuse, experimentation from exfiltration, and productive AI adoption from hidden risk.

Figure 1: Darktrace / SECURE AI reconstructs the full sequence of events, showing every user and agent interaction in context, with risky prompts highlighted and categorized, including PII, sensitive data, and other policy violations.

At Darktrace, this is the key lesson emerging from the market: prompt language does matter, but it does not stand alone. It is most valuable when treated as a new behavioral input that can enrich understanding across the enterprise, not as a self-contained source of truth.

Why prompts become less useful when analyzed in isolation

The curious case of prompt language analysis, then, is this: the more important prompts become, the less useful they are in a vacuum.

The real opportunity is not just to see what was asked. It is to understand why it was asked, what it meant in that moment, and what happened next.

For a deeper look at how organizations are approaching this challenge from the strengths of prompt analysis to its limitations in isolation see Prompt Security in Enterprise AI: Strengths, Weaknesses, and Common Approaches, which expands on the role prompt-level controls play within a broader, context-driven security strategy.

Continue reading
About the author
Nabil Zoldjalali
VP, Field CISO
あなたのデータ × DarktraceのAI
唯一無二のDarktrace AIで、ネットワークセキュリティを次の次元へ