In an industrial network, malicious activity can be hard to tell apart from normal data exchange between devices. Malicious actors routinely rely on legitimate tools and the Living-off-the-Land technique, so a single network connection may show no known signs of an attack: the device uses a familiar protocol and transmits ordinary data. An anomaly can hide in subtle details: the data source, the time of access, or the volume of transferred data may be atypical for a particular device. However, to detect this kind of anomaly, you need to understand the roles of the devices at both ends of the connection and know which network activity patterns are typical for them.
But an organization’s asset inventory data is often incomplete, which makes it harder to catch unauthorized connections in time. Suppose monitoring detects a host that regularly exchanges data with programmable logic controllers (PLCs). Its IP address is known, but the inventory does not specify what role it plays: is it an engineering workstation, a process data server, or a piece of contractor’s equipment? The host may also have been on the industrial network for a long time, with its record going out of date after a network upgrade.
In the joint OT asset inventory guidance released by CISA and its partners in 2025, inventorying devices and classifying them by role are treated as part of OT asset management. For organizations, this is one of the most important areas: according to the SANS 2025 ICS security survey, 50% of respondents had deployed or significantly expanded ICS inventory and visibility tools over the previous 12 months, and 54% named them as investment priorities for the next 12–24 months.
In Kaspersky Industrial CyberSecurity for Networks, the Asset Discovery technology is the foundation of asset visibility. Passive network traffic analysis detects devices and their interactions, while recognition rules help determine the device category and extract available information about it, such as the vendor, model, or software version. Exactly which attributes the system can obtain depends on the protocol and the traffic it observes. We covered this in more detail at the Kaspersky Industrial Cybersecurity Conference.
Hardware and software information in the device card
The product also supports active polling modules (APMs): where polling is allowed, they can request additional configuration and firmware version details from devices. This kind of inventory is far more detailed than a list of IP addresses.
Installed modules and their placement in a PLC rack
However, even accurate hardware information doesn’t always explain what function a host performs in a specific industrial facility. Gaps appear for various reasons. Some devices send identifying information only at startup or during configuration, while others rarely communicate. Proprietary protocols can’t always be parsed, and encryption may leave only connection metadata for analysis. If the monitoring system doesn’t receive traffic from every segment of the industrial network, some network activity goes unobserved. Active polling doesn’t suit every device or every operating condition.
Devices of unknown category
When there isn’t enough direct information about a device, its category can be refined by looking at the patterns of its recurring network interactions. Analyzing these interactions is the job of Machine Learning Event Processor (MLEP), a service included in Kaspersky Industrial CyberSecurity for Networks 5.0. This machine-learning-based event processor can both refine device categories and detect anomalies in their network behavior.
To compare device behavior, you don’t need to store every packet as a separate observation (a data object) in the ML model. MLEP represents network sessions as a sequence of short 15-second fragments. For each fragment, MLEP stores information about the endpoints (IP, MAC, and port), the direction of data exchange, the protocol, and the number and size of the packets transmitted. Together these make up an event, which then gets a timestamp. Traffic thus becomes a sequence of events showing which hosts a device communicated with, when, and how communication intensity changed.
Aggregating data into events also sets the method’s limits. An aggregated fragment can’t be used to reconstruct the exact order of individual packets or the content of commands. MLEP can draw conclusions about a new protocol only if another KICS technology has managed to recognize it. That said, comparing stable network connections and changes in exchange volume doesn’t require a detailed record of every packet. These observations form the basis of the device’s behavioral history.
Each object has its own behavioral history, and one object’s history can’t be applied to another. But the principles of behavior analysis, along with some common rules, work across different objects.
Let’s look at how this works in MLEP, which is built on a spiking neural semantic network (NSN).
Neurosemantic networks structure information as a multilayer ordered graph, where the data in each consecutive layer builds on the data in the previous one. The nodes of this graph — that is, the units of information in each layer — act as neurons.The NSN creates dedicated channels to handle different event parameters (port, IP, and so on). Unique parameter values for each channel are stored as neurons in layer 0. The event itself, which consists of a set of parameter values, is stored as a neuron in layer 1 of the network. If an event occurs for the first time, neurogenesis creates a new neuron with input connections to the corresponding parameter neurons. This lets the network learn and operate on the event stream at the same time. If the event has occurred before, the network activates the same neuron that was created when the event first appeared. This way, the network identifies each event precisely. Each neuron that corresponds to an event also collects basic statistics: when it was last activated and how many times it has been activated.
In NSN-based event processing, layer 2 and above store sequences of events over time: recurring patterns and the full history. However, categorization and statistical anomaly detection don’t require storing the full packet sequence or a detailed timeline of every network event. As we noted earlier, events are themselves aggregates, which means the order of network packets has already been averaged out. So MLEP keeps history only temporarily, within each episode (30,000 aggregated events by default), which saves computing resources.
MLEP processes statistics mainly through specialized listener neurons on layer −1 (minus one) of the network. Attention heads and monitors are built on these neurons.
Key components of the NSN
When an event arrives, the corresponding neuron is activated, and the network instantly knows whether that neuron is new or was created earlier. Each attention head subscribes to the events or event features it needs and takes in only the activations of the matching neurons. If an event occurs for the first time, all heads are notified and then decide whether to subscribe (grow a connection to the relevant neuron) and receive information about activations of such neurons from then on.
It is called a spiking network because activation information is a spike, not a number. This spike tells the head that something it’s interested in has happened.
An attention head is configured according to scenarios — the situations that MLEP should send alerts about. The core of a head’s configuration is a query, which filters incoming information and serves as the subscription criterion. Unlike regular database queries, this query runs on the stream, so it can react to an event instantly.
Attention heads work very much like the semantics of ordinary human language. A query has a subject of attention (one or more grammatical subjects) and conditions (the other parts of the sentence). In the infrastructure of an industrial enterprise, the subjects are devices on the network, and the conditions are various aspects of network communication, such as ports, protocol, and average packet size. If a head has many subjects (for example, if it’s interested in every device on an industrial network), it fans out: it creates a child head for each individual value of the subject of attention, that is, each device. The child head then watches a specific device and reports to the parent head.
Attention heads don’t accumulate or process statistics. They only filter event streams. When an attention head receives a relevant event, it activates and sends a spike to the monitor subscribed to that head.
Monitors, in turn, have stacks for accumulating and processing events. They can also have their own queries and post-filter information from the head. For example, they can select only new events from those filtered by the head.
There are several types of monitors:
MLEP uses these monitor types to implement scenarios for categorizing devices and detecting their anomalous behavior:
Groups of devices with similar network behavior
At the final stage, the system verifies that assigning the category to the asset is permissible. This is again done by special rule monitors loaded into the NSN at system startup. The rules rely on asset metadata available from Asset Discovery and APMs.
The current version of MLEP ships with preconfigured heads and monitors.
The functional dynamic diagram below shows how events from devices on an industrial network are processed.
Event processing flow (click to play)
At a facility with several similar production lines, device information may be more complete on some lines than on others. On one line, devices are identified as PLCs, while on another, after an upgrade, some hosts show up only as addresses: the available traffic didn’t provide enough information to determine their category. Still, these devices communicate with the same types of HMIs and process data collection servers. Their communication directions, frequency, and volumes may be similar.
That’s where MLEP comes in. It takes the categories of already known assets and matches recurring features of their network behavior against those of uncategorized devices. When the observed interactions turn out to be similar, MLEP can add a category for that host to the inventory.
Device category suggested by MLEP
Once the category is refined, the industrial network map provides more context about the equipment on the network. When analyzing network activity, the analyst immediately sees the device’s suggested category and can skip looking up each IP address in the design diagrams. This approach is especially useful where equipment changes faster than the documentation is updated.
The suggested category is also checked against existing device information and defined constraints. This helps filter out some contradictory options, but prediction quality still depends on how much of the traffic is visible, especially if the device communicates only rarely.
Investigations of real-world attacks show that attackers actively use both standard tools and permitted protocols. You don’t always need to know an attacker’s specific technique in advance to spot potentially malicious activity. Signature-based detection helps find known signs of an attack, but it doesn’t assess how typical an observed interaction is for a particular device. MLEP compares current activity with the device’s usual network profile: which hosts it communicates with, which protocols it uses, when and how much data it transmits, its average packet size, and so on. The changes it detects help guide the direction of the investigation. Correlating these changes with other events and the operational context helps determine whether malicious actions could be behind them. When investigating an incident that has already been detected, anomalies help reconstruct the sequence of actions.
The first version of MLEP comes with six preconfigured monitors, one for each type of deviation from the device’s usual network profile:
| Anomaly | How it helps with event analysis |
| New external host | Flags a new external connection that wasn’t previously seen in the device’s observed behavior, even if the IP address isn’t listed among known IoCs. Helps check whether this connection appeared at the same time as other suspicious events. |
| New combination of destination address and port | Shows that the device has begun using a service it hasn’t used before. Once an incident has been detected, this helps trace new connections after the more critical events have been analyzed. |
| Unusually large volume of data sent | Highlights data transfers that are atypical for the device and helps correlate their timing and direction with other events. |
| Atypical activity time | Shows that the device was active at a time that’s unusual for it. In an already detected incident, this helps reconstruct the timeline of actions around the main alert. |
| New protocol | Helps spot a new method of communication, even when the device is communicating with familiar hosts. |
| Activity with signs of scanning | Shows that the device has started contacting an unusually large number of neighboring hosts. This signal helps reveal possible reconnaissance ahead of attempts to access equipment. |
During an investigation, anomalies help reconstruct the sequence of changes: when a specific host established new connections, switched protocols, increased its data transfer volume, or shifted its operating hours. Matching events by device and time shows an analyst how activity unfolded around an incident that has already been detected.
This is especially useful when some of the attacker’s actions fell outside the monitored scope or didn’t trigger other KICS technologies. Correlating these changes with the device’s role, protocol analysis data, and events from other technologies gives the analyst a more complete timeline. Behavioral anomalies show exactly where to look for the missing links and which network interactions need attention.
Atypical Activity Time anomaly card
In the event list, anomalies detected by MLEP are marked with an AI label. Each event shows its name, source, destination, timestamp, and the ID of the triggered scenario.
In an industrial network, whether behavior is anomalous depends on the host where it occurs and that host’s place in the industrial process. The repetitive nature of many communications makes a change in normal data exchange easier to spot, while knowing a device’s function and criticality helps gauge the possible consequences. Taken together, these observations make network traffic more than a list of connections: it helps assess new events in the context of a specific facility, even without an attack scenario known in advance. Machine learning technologies are well suited to this kind of comprehensive analysis, as they can process and organize huge streams of heterogeneous data. Starting with version 5.0, Kaspersky Industrial CyberSecurity for Networks handles this function with MLEP, which works effectively alongside the product’s existing functionality.
A modern organization protects its information systems with a whole suite of technologies that generate thousands of security alerts every day. Looking at these alerts in bulk, you’ll see that a significant share are false positives. Ideally, these alerts would be filtered out automatically so that analysts could focus on the ones that may point to a real incident. However, writing filters is a time-consuming task in itself. To create a correct filter, an analyst needs to:
This post looks at how filter creation itself can be automated to reduce analyst workload.
In the context of security alerts, a filter is a combination of event field values that marks a detection as a false positive and not worth the analyst’s attention. Filters can be generated for any combination of event fields, but not every combination yields a valid filter. That’s why we need to define validity criteria: conditions showing that a specific filter actually removes noise and doesn’t suppress useful alerts.
Detection rules in security systems typically contain a set of conditions, and each of them can generate an alert. A specific event triggers a given fragment of a detection rule. The set of events that a given fragment responds to is what we’ll call its context. A good filter should cover the entire context where possible, without going beyond it. Filters that are too broad or too narrow are useless. For example, a filter based solely on PowerShell would exclude too many events, including some that deserve the analyst’s attention. Most of the filtered events would also be unrelated to the rule in question. A filter on a specific ID, on the other hand, would cover only one event, which may never recur.
We can run into a number of problems when creating filters. First, events often contain an excessive number of fields (around 400 on average), and most of them aren’t needed for filtering. At the same time, generating and ranking combinations of these fields is also fairly resource-intensive. So, the first step is to discard the irrelevant fields and keep only those worth using in the filter. For example, out of 400 fields in an event, four may be enough for filtering.
Another problem is string normalization. Field values (for example, the path to the same application on different systems) can vary depending on system settings, user folder names, and so on. For a filter to account for all these variations, the strings must be normalized. For file paths, in particular, you can replace the drive letter and user folders with standard environment variables and convert all paths to a single case.
Finally, if a filter completely mirrors the query of a detection rule, it effectively disables that rule. To fix this, remove the event fields used in the detection rule from the filter. For example, if a rule triggers on a file path, there are two ways to filter out unwanted alerts:
Entrar
file_path: ".../something.exe", you can set up a filter that excludes events with the user/unique_files/ directory in the file path.Given the problems described above, automatically generating filters for security alerts should consist of the following steps:
For example, consider an event described by the following JSON:
After generating and ranking the combinations, we got three filters based on file paths:
processcmdline:("nodejs\\node.exe" AND "*****\\appdata\\roaming\\npm\/node_modules\/pi-coding-agent\/dist\/cli.js")
file_path:"*****/.pi\\agent\\bin\\fd.exe"
processcmdline: "programfiles\\nodejs\\node.exe"
The last filter turned out to be too broad, but the analyst marked the other two as valid.
The algorithmic approach to filter generation has several advantages. First, it’s simple and deterministic: the same alert always produces the same result. Second, it’s interpretable, meaning we can see why a given combination ranked near the top of the list and why a particular field didn’t make it into the list of candidate fields at all. Third, the stages of the process are independent of each other, so individual steps can be improved and tuned without the risk of breaking the algorithm elsewhere. Finally, it has no external dependencies or models, which makes it easy to deploy in any customer environment.
However, it also has drawbacks. The first is its dependence on context. If there’s no context for some reason — for example, the rule’s first-ever trigger — a filter can’t be created. The second problem is string normalization. There’s a high chance of running into atypical cases that the normalization rules don’t cover. Often such cases can’t be described correctly at all, because they contradict each other.
These problems can be solved with large language models (LLMs). When an LLM generates the filters, no context is needed, because the LLM doesn’t iterate over or rank combinations. The LLM also takes a more flexible approach to string normalization, including non-standard cases. That said, an LLM doesn’t have to be involved at every stage of the generation process. We decided to keep the algorithmic solution where it works well. For example, a simple algorithm selects suitable fields without errors, so we didn’t move this stage to the model.
The LLM-based filter generation looks like this:
The input is still an alert containing an event with 400 fields, plus the rule that triggered on that event. Then an algorithm selects the most suitable fields for the filter. Next, the LLM processes the input — the event, the selected fields, and the detection rule — and generates three filter options along with an explanation for them. The explanation helps the analyst understand why these particular filters were suggested and make an informed choice.
Consider the following event in JSON format:
The LLM suggested three filter options that all contain the same command line fragment:
file_path: "\\WindowsPowerShell\\powershell.exe" AND filecmdline: "Get-VM -Name kafka\*"
filecmdline: "Get-VM -Name kafka\*"
filecmdline: "kafka*"
As in the previous example, two of the three output filters were valid. The last one was too broad to be usable. The analyst chose the first filter with minor edits:
file_path: "powershell.exe" AND filecmdline: "Get-VM -Name kafka\*"
Using an LLM solved the problems of purely algorithmic filter generation: dependence on context and errors in string normalization. However, the single-request approach had its own drawbacks. First, it had no output validation. The LLM returned filters as-is, with no syntax check and no way to correct errors. Second, unlike in the algorithmic version, all filter generation tasks were handled by a single system prompt. As a result, a minor fix in one place could break the logic in another.
To address the problems of generating filters with a single model and a single system prompt, we tried distributing the tasks across different LLM agents. In this version, we brought in agents as early as the stage of selecting relevant fields. The chain includes four agents in total:
For the following JSON event, the agents produce three valid filters based on the file path.
The filters differ in the additional parameters that support the detection:
file_path:"sysvol\\domain\\scripts\\logon.bat"
file_path:"sysvol\\domain\\scripts\\logon.bat" AND file_md5:"0x7..6E"
autorun_entry: "Group Policy\\State" AND file_path: "SYSVOL\\domain\\scripts\\logon.bat"
LLM agents give us a system that can validate and correct filters. Its modular architecture also lets us tune and improve different stages of generation independently. However, this approach makes more LLM calls than the previous ones, so the system is slower. It also depends on access to a model, which may not be available in the organization’s isolated environment.
To evaluate how well the filter generation system works, we use two main metrics:
Below is a comparison of the three filter generation approaches by these metrics:
As the results show, LLMs significantly improve the efficiency of filter generation. At the same time, the LLM agentless and agent-based approaches perform almost identically. So we chose the simpler and faster option: an LLM with a single system prompt. As a result, we cut false positives by 99%, increased the number of filters generated per day by 28%, and made analysts’ alert handling 17% more efficient.
Practice has shown that filter generation can be automated effectively. We’ve looked at several options for automatically generating filters for false-positive security alerts: a purely algorithmic approach, an LLM-based approach, and an approach built on a group of LLM agents. All of them can generate valid filters and make the analyst’s work much easier, but each has its own pros and cons.
The algorithmic approach scored lower on the evaluation metrics, but in two-thirds of cases, it still produces filters that satisfy the analyst. It works especially well when there’s already a set of similar events to filter out, meaning enough context has accumulated. The algorithmic generator can also be deployed in any environment. Both standalone LLM and multi-agent LLM approaches show high accuracy, and their results differ only slightly, so the slower agent-based approach looks excessive at this stage. If an LLM is available, the best option appears to be a single model that handles several stages of the generation cycle.
AI agents and associated tooling are becoming increasingly common in corporate infrastructure. Agents can perform a range of actions with the user’s privileges, such as accessing data, running commands, and interacting with external services. This creates new security risks, especially when these tools are used without the security team’s oversight or well-designed policies.
Detecting, tracking, and controlling AI agents is becoming one of the most pressing tasks for security teams today. We closely follow technological developments and continuously update our solutions to respond to new threats in a timely manner amid a rapidly changing technology landscape. This post describes new expertise built into Kaspersky products that helps counter the threats posed by the proliferation of AI agents. We approach this problem from several angles.
First, you need to thoroughly understand the threat you are dealing with. To this end, we added profiles to the Kaspersky Threat Intelligence Portal that describe various AI agents and how attackers exploit them. At their core, AI agents are programs (processes) that run in the operating system on endpoints and can also launch other processes. This means that Endpoint Detection and Response solutions are among the key endpoint controls for this threat. These solutions detect the presence and launch of AI agents, track their activity, and identify potentially dangerous behavior. When needed, they also take action, such as stopping a suspicious process, blocking the activity, or isolating the host, to reduce the likelihood of an incident escalating. For this purpose, we developed specially adapted detection rules.
You can use Kaspersky Endpoint Security, now also available as Kaspersky Next, to control and restrict the launch of AI agents. This solution uses two components to detect AI agents and control how they are used. Application Control categorizes the application processes launched on endpoints, and Web Control manages access to specific categories of web resources for users on endpoints.
Finally, even harmless AI agents can contain malicious instructions (skills), such as those downloaded from unsafe external sources. Kaspersky Scan Engine analyzes installed applications to detect these threats and also checks LLM files for malicious code. Let’s take a closer look at these solutions and technologies.
Tools from Anthropic and OpenAI take the top spots in the September 2026 ranking of AI agents by popularity among Kaspersky users. This list is based on telemetry from Kaspersky solutions. It shows the number of users who had the relevant AI vendors’ products installed at the time of the study.
| # | AI product |
| 1 | Claude |
| 2 | Claude Code |
| 3 | Codex |
| 4 | OpenCode |
| 5 | Zed |
| 6 | Wispr Flow |
| 7 | LM Studio |
| 8 | ZCode |
| 9 | Chatbox |
| 10 | Cherry Studio |
The data shows that agentic systems, especially AI agents for software development, are the most popular AI applications. The TOP 10 also includes an application for running large language models locally (LM Studio) and a voice note-taking app (Wispr Flow).
According to open sources, these are the most widely used AI services in 2026:
| # | AI service |
| 1 | ChatGPT |
| 2 | Google Gemini |
| 3 | Claude |
| 4 | Grammarly |
| 5 | DeepSeek |
While analyzing the most popular and widely used apps, our experts also noticed a sharp rise in vulnerable versions of AI-related software and services. From 2025 to date, more than 2500 vulnerabilities have been published:
Number of published vulnerabilities in LLMs, AI tools, and plugins with similar functionality, 2025–2026 (download)
* Data for previous quarters was updated and may differ slightly from previously published data.
To enforce the organization’s security policy for AI agents and other AI applications on the corporate network, Kaspersky products offer application control based on KL categories. These categories classify installed software and give you granular control over which approved applications can run in the organization.
Application Control analyzes the processes launched on endpoints, tracking which AI agents are running in corporate systems at any given moment, on which hosts, and under which accounts. Application Control also shows how long a given program has been in use.
This makes it easier for the Kaspersky Security Center (KSC) administrator to control applications. Administrators can use KL categories to allow or block AI agents inside the company for specific groups of employees.
Here is what this functionality looks like to KSC administrators:
KL categories available when configuring Application Control
For details on how to configure Application Control in KSC, follow this link.
In Kaspersky Endpoint Security (KES), Application Control can operate in two modes:
Below are the steps to configure Application Control to block AI tools.
Configuring Application Control for the AI Tools category in KES
With this setup, when a user tries to launch an AI agent, they get a notification and the agent doesn’t start.
KES notification about an attempt to launch an AI agent
For more on how to configure Application Control in KES, follow this link. Our public portal also lets you explore information on KL categories and the applications that fall under them, in particular details on how AI agents are categorized and how popular they are.
AI Tools KL category
The View applications drop-down list shows products for a specific vendor.
Similarly, Web Control lets you manage user access to web resources. If a user tries to open a denylisted website, KES blocks access and shows a warning.
KES notification about an attempt to access a prohibited web resource
When configuring access, the administrator can block websites that provide AI as a service.
Website categories
For more on configuring Web Control, follow this link. The help portal also provides information on website categories that can be assigned to resources to configure access control.
To an EDR solution, an AI agent is a regular process. It has its own executable, a tree of child processes, and network communications, all of which need to be monitored for suspicious activity. An attacker can use an AI agent as an automation tool. An AI agent installed as a legitimate user tool can also be leveraged in a ClickFix attack, where it executes a malicious instruction received from an untrusted source, such as a document, web page, or email. That said, unsafe actions by an agent do not always indicate an attack. There may be no malicious intent in the chain at all. An incident can result from what looks like an ordinary instruction from a user who doesn’t fully understand the consequences of having it executed. The mere fact that an agent was launched does not reveal whether it is being used in an attack. That is why Kaspersky EDR (KEDR) uses both conventional behavioral rules and rules written and adapted to detect AI tools. The latter identify risky agent actions and settings, and help determine which AI tool was involved in suspicious activity.
Conventional KEDR rules detect the vast majority of familiar malicious techniques, including when performed by an AI agent. A suspicious PowerShell launch, persistence via Task Scheduler, or an attempt to steal credentials leaves clear, identifiable traces. These traces look the same no matter what initiated the actions. The rules analyze launch parameters, process relationships, affected files, network activity, and other telemetry events.
For example, in a recent Armored Likho campaign, we found signs that AI had been used to create loader and dropper samples, which were later compiled with NSIS (Nullsoft Scriptable Install System). The key telltale artifacts were icons in the scripts and excessive comments on every function, which is uncharacteristic of human-written code. In addition, the group continues to generate new modules for its BusySnake Stealer and BusySnake RAT samples. For example, the attackers used AI to develop a module that runs multiple checks for virtual environment artifacts and calculates a score. Depending on the score, the sample either runs or exits with code 0. Even though the initial access samples, payloads, and their modules were AI-generated, the commands they execute via rundll32 and PowerShell are detected by the shell_creation_by_rundll32 and suspicious_powershell_cmd_or_script_spawning rules.
So, for detecting an attack like this, it is not important who wrote the loader, the payload, or its modules: they are given away by the commands they execute, process relationships, and other traces of activity in the system. AI can be involved both in preparing the code and in carrying out the attack directly on the device.
In the latter case, the attacker uses the agent to carry out their tasks and lets it choose and run the required commands within the privileges it has. For effective protection, it is critical to link suspicious actions back to the AI tool and the conditions under which it operates. In KEDR, several specialized rules handle this task.
We have created separate rules for popular AI agents, ai_agent_*_access_to_sensitive_file_**, where * is the tool name and ** is the operating system. They generate alerts when an agent accesses files that contain credentials, tokens, or keys.
The ai_agent_dangerous_execution_mode rule detects agents launched with auto-approval of all actions (so-called YOLO mode) or with the sandbox disabled. These settings do not indicate an attack on their own, but they let the agent perform dangerous actions without user confirmation.
The suspicious_mcp_server_artifact rule triggers when MCP server scripts are created or modified in hidden, temporary, or cache directories. Attackers can place an MCP server in these locations to quietly connect it to the agent: through the MCP server, they can pass instructions to the agent, call third-party tools, and access the user’s data.
The llm_api_access_by_unusual_apps rule generates an alert when an application on Windows, Linux, or macOS connects to the API of a known external LLM service, unless this activity falls under predefined exclusions. This helps spot model access by applications that do not normally behave this way.
Some Kaspersky EDR rules tag events for investigation but do not generate alerts. For example, the ai_agent_**_execution rules flag launches of known agents on Windows, Linux, and macOS. The local_ai_model_artifact rule records the creation or modification of AI model files in typical application, runtime, and cache directories. These events reveal AI tools on the device, including tools that run on local models and never call an external API.
The Kaspersky Threat Intelligence Portal offers a large number of software profiles, including profiles of AI agents. You can sort them by the ai_ prefix:
AI agent profiles on Kaspersky TIP
The profile of an AI agent such as Claude Desktop includes a detailed description of the software, the privilege level required to run it, supported platforms, and threat actor groups known to use it. It also lists tactics and techniques mapped to the MITRE ATT&CK framework, with descriptions of the procedures the tool can be used for.
Claude profile on Kaspersky TIP
The description also lists ways to detect the software through standard system auditing mechanisms:
The detection section lists the main artifacts the software leaves in the system: command lines, possible launch parameters, names of the services and scheduled tasks it creates, and the key network artifacts.
Detection section in the Claude profile
In December 2025, the Kaspersky Threat Intelligence Portal added Hunt Hub (HH), a repository of detection rules. This is a unique Kaspersky project, and we developed it in response to growing market demand.
This functionality is available in the Threat Landscape section of the portal. It gives analysts access to detection rules: their descriptions and their mapping to MITRE ATT&CK tactics and techniques, as well as to the associated cybercriminal groups.
In the Hunt Hub section, customers can explore the detection logic of EDR, NDR, and NGFW products, including the logic for detecting AI agents.
Rules in the Hunt Hub repository
The rule card provides:
Detection rule card in Hunt Hub
Skills are a key component of the AI agent ecosystem. They’re prewritten sets of natural-language instructions that extend agent functionality. We previously covered the risks that malicious skills pose to agent users. A skill can require the usage of additional programs and services, come with helper scripts (for example, written in Python), or contain a malicious command embedded right in the instruction text. This means that adding skills to an originally harmless AI agent opens up several new ways for malicious code to enter the system. To prevent this, any skills obtained from external sources must be scanned. Kaspersky Scan Engine is well suited for this task. This is a cyberthreat detection and response solution that seamlessly integrates with various applications. It can detect malicious code in agent skills, associated scripts, and in files generated by running agents.
Detection of a malicious agent skill in Kaspersky Scan Engine
Detection of an LLM with malicious code in Kaspersky Scan Engine
AI agents are rapidly making their way into corporate environments: they help employees work faster and take on a wide range of tasks, but they also open up new opportunities for attackers. To take advantage of this technology without compromising security, organizations need to build multi-layered protection and minimize risks. A range of Kaspersky solutions can help with this:
When companies combine the tools described above in a comprehensive approach, they can adopt AI agents in an informed way: with an understanding of the associated risks and effective means of minimizing them.
Imagine you’ve ordered a small paper unicorn from an online marketplace. You clicked “Buy”, received a cheerful notification, and expect a courier to bring the box tomorrow. If the unicorn is delayed by a day, it will be annoying but not catastrophic: an apology, a discount promo code — and the incident is forgotten.
Now let’s multiply one late unicorn by several hundred thousand orders.
Packages start heading to the wrong cities. The website shows an item is in stock, even though the required bin is empty. Conveyors stop, boxes get stuck between floors, the cargo piles up with no freight manifest. Trucks line up at the gates, and every minute of downtime turns into blown schedules, support tickets, fines, and lost sales.
It is at exactly this scale that a cyberincident in a warehouse becomes a problem for the entire business. A modern warehouse is no longer just a big building with shelving. It is a tightly interconnected hybrid system: web applications, radio channels, and an automated factory. Physical processes here depend on data, and errors in data have very physical consequences. Examples include an outage that affected warehouse management systems at the grocery giant Morrisons, and the 2025 Form 10-K annual report of United Natural Foods, Inc. (UNFI), which confirmed unauthorized activity on information technology systems that impacted order fulfillment. More recently, reports emerged that a cyberattack disrupted eight CEVA Logistics warehouses across Europe and exposed customer delivery data.
This paper follows a single order through a warehouse to show where the operation is vulnerable, the stages it traverses, the systems that drive it, and the points where an attacker can reach them. We then set out the threat model: what is at risk, who is in a position to take it, and how each risk lands on the business. From there, we look at how an attacker gets in, and what reaching the warehouse management system costs in practical terms: orders unshipped, inventory records nobody can trust, and an operation that stops. We close with recommendations on where the controls that matter actually sit.
A modern warehouse is a sequence of stages involving many machines and people, all working together to keep the business process stable so that, in the end, your order gets delivered. This process consists of eight main stages, shown in the figure below.
Stages of order fulfillment
So let’s assume we’ve ordered our paper unicorn and are waiting for it at home. Let’s follow our order through all the main stages, from its arrival at the warehouse to the final stop on its journey.
All eight stages are coordinated by the WMS. It looks like a large web application, but in its role it’s closer to the operating system of the entire warehouse: it knows which goods exist, where they’re located, who or what should move them, and where they go next, linking employees, inventory levels, storage addresses, tasks, and orders. It also exchanges data externally: with ERP systems, transport platforms, marketplaces, dark stores, and external carriers.
Surrounding the WMS is an even larger digital ecosystem: identity services, databases, APIs, printers, carrier systems, technicians’ laptops, video surveillance, access control, fire and gas alarms, power supply, and a building management system (BMS).
That is why a single “warehouse network” doesn’t really exist. As we’ve seen, there are many overlapping technologies — IT infrastructure, OT systems, radio, cloud services, building security, and automation — layered on top of one another. A lone vulnerability scanner will never tell you whether a warehouse is secure.
Securing an environment such as a warehouse requires a comprehensive security assessment covering all of its interconnected systems. This involves identifying vulnerabilities across each system and understanding how weaknesses can be combined into attack paths that a real attacker could exploit. But a vulnerability only matters in proportion to what it exposes and who is able to reach it, so before looking at the weaknesses themselves, we need to understand what an attacker in this environment is actually after, because in a warehouse, the blast radius extends well beyond data. Everything we just traced — the systems, the devices, and the people operating them — exists to keep goods moving, and attacks in various parts along that chain can affect the business differently. The chart below shows what is at risk in the warehouse and where those risks land for the business. The segment on the left groups them under the three core security properties: integrity, availability, and confidentiality. The segment on the right maps each property to the business outcomes a failure can cause.
Warehouse risks
The first area at risk is integrity. Inventory counts, locations, orders, and shipment data must stay in sync with the physical world. Corrupting a record silently desynchronizes the operation, and because the eight stages form a dependency chain, one falsified entry early on is inherited downstream, surfacing later as stockouts and mis-shipments. In its deliberate forms, this lets an attacker redirect goods, release stock, and perform other actions on behalf of the platform’s vendors under cover of legitimate records. Integrity is also a safety concern: when people or machines act on information that no longer reflects the physical environment, a data failure can become a physical incident. This is why integrity is the only property linked to all five outcomes, including harm to people.
The second is availability, the warehouse’s ability to keep goods moving. Halting a conveyor, a sorter, label printing, or the WMS stops orders from shipping, and in a business built on speed, every hour of downtime means delayed deliveries and broken promises. Ransomware can bring the whole warehouse to a standstill. The impact shows up as operational disruption, lost revenue, and reputational damage.
The third is confidentiality. Customer data, business records, and system credentials are valuable targets. Their exposure leads to legal liability and reputational damage, and stolen credentials often open the door to further attacks. Where the WMS connects to financial or ERP processes, the risk extends to theft of funds.
Knowing what is at risk tells us what an attacker stands to gain; the next question is who is in a position to take it. Threat actors differ in the access they already hold. For example, an attacker may be authenticated and privileged, have limited physical or remote access, or have no legitimate access to the environment at all. Based on their relationship with the warehouse and their level of access, threat actors can generally be divided into two categories:
Combining these threat actors and their access levels with the risks identified earlier, the vulnerabilities present in each system, the resulting attack surfaces, and the potential attack paths between them provides a threat model for the warehouse environment. This allows an organization to understand how different systems and people could be targeted, how individual weaknesses could be combined, and where security controls are most needed.
Knowing what is at risk and who might come after it still leaves the most practical question open: how they would actually get in, and how far they could go once they did. The diagram below sets out the warehouse’s attack surface: every point where an attacker can get in, organized by the access that reaching each point requires.
Warehouse attack surface
On the left are the attacker categories discussed above, distinguished by where they stand and what access they already hold. Each reaches the environment through a different set of entry points, and the impact of a successful attack can land on several systems, from the WMS to OT systems such as robots, sorters, and physical access control system (PACS). The WMS is highlighted because it is the most significant of these: it coordinates the OT systems, so compromising it can affect the entire operation rather than a single process.
Next, we will cover the initial access vectors, the points a threat actor can use to reach the warehouse environment. Following that, we turn to the impact on the logistics processes themselves, focusing on the WMS, since its compromise affects every logistics operation that depends on it.
Remote access through internet-facing services, third-party connections, and corporate networks gives an external attacker a foothold on the warehouse network infrastructure. It is a well-covered problem and not specific to this environment, so we note it here as an entry point but will not examine it further.
Wi-Fi connects handhelds, tablets, printers, cameras, and workstations. Bluetooth serves scanners, headsets, and positioning beacons. Zigbee and similar low-power protocols run lighting, sensors, and building automation, while LoRaWAN carries telemetry over long distances. Cellular provides a backup channel for kiosks and contractor equipment, and proprietary protocols control industrial panels and specialized machinery.
Radio passes through windows, gates, roofing, and thin external panels, so a network built for receiving and processing goods may be reachable from the parking lot and, with a directional antenna, from considerably farther away.
That reach matters in two ways. It gives an external attacker a place to start, as proximity alone is enough to reach the Wi-Fi and the low-power protocols, and from there to gain an initial foothold in the warehouse networks.
However, radio does not only lead into the network. It also connects directly to the field devices and carries the live control traffic of warehouse operations, which makes availability a first-order concern. If hundreds of handhelds lose their connection at once, picking stops. If an AGV loses its control channel, the safe reaction may be to halt in place, and a stopped machine can block an entire route.
The scale of the warehouse itself is part of the attack surface. A fulfillment warehouse can run up to tens of thousands of square meters, with racking climbing several stories and walkways threading through it. Structured cabling runs along that racking, cameras and access points are mounted on that, and wall jacks sit behind pallets that have not moved in months. Almost none of it is in anyone’s line of sight.
That gives a person in the warehouse something a normal office does not: unsupervised access to live network infrastructure. Unplugging a ceiling camera leaves an active port. An unused jack behind a shelf leaves another. Neither requires tools, credentials, or any interaction with a system that logs anything. And since the camera or access point sits on the same segment as the rest of the site, the compromised port, not being isolated, opens a path to the network core.
Barcode scanners are everywhere in a warehouse. They are the main tool workers use to tell the system what is moving and where, supporting receiving, picking, packing, and inventory management. A typical setup has handheld scanners communicating with a base station, which is in turn connected to a workstation or another business system.
This makes the scanner an entry point into the core network in two ways.
First, through the barcode itself. Since the scanner acts as a keyboard to the connected host, a maliciously crafted barcode or QR code can inject keystrokes when scanned, potentially giving the attacker an initial foothold on the host without directly accessing the network.
Second, through the scanner’s wireless link. During a customer engagement, with the provided test stand of such a device, we discovered that an attacker within range could inject keystrokes into the connected host and potentially achieve remote code execution without user interaction. This attack pattern was observed across multiple vendors, including, for example, Honeywell devices affected by CVE-2026-4272.
In both cases, the result is the same: a foothold on a host inside the warehouse network. The scanner is simply the entry point; the significance of that foothold comes from the host’s trusted position within the operational environment and its connectivity to core systems.
Portable data terminals (PDTs) are essentially ordinary Android devices with an embedded barcode scanner, carried by workers to scan items; each scan is passed to a dedicated application that forwards it to a remote server over an API. To keep workers on task, the devices usually run in kiosk mode, locking the user to that single application and preventing interaction with the rest of the Android system.
That lockdown is exactly what an attacker wants to break, and breaking it is what turns the terminal into an initial access vector: escaping the kiosk can expose the Wi-Fi password of the network the terminal is connected to, or allow malicious software to be installed that gives the attacker remote access to the network over the internet.
Warehouses contain different types of printers. Alongside the usual IT printers used for documents and paperwork, there are specialized thermal printers, small devices that use heated elements on the printhead to create labels, barcodes, and tags. As we mentioned earlier, these are mainly used to print the labels, usually containing barcodes, that are attached to goods during the putaway stage.
Here we focus on mobile thermal printers: small, battery-powered devices that workers can carry or wear. They typically connect to the internal network over Wi-Fi to receive print commands, while also supporting Bluetooth for connecting to nearby devices. This combination makes them a potential entry point into the internal network.
Some models, such as the Zebra ZQ630, support multiple interfaces and can run applications directly on the device. During a security assessment, we discovered that apps can be uploaded and executed over Bluetooth. By exploiting a pairing misconfiguration, an attacker within Bluetooth range could bypass pairing, upload malicious software, and gain an initial foothold through the printer.
With a foothold in the network, an attacker can pivot to the logistics processes themselves: from the WMS through the warehouse infrastructure to the PLCs and the machinery they drive. As we saw earlier, all OT systems and devices are interconnected and the WMS sits at the head of that chain: it is the system of record every other process reads from and writes to, so compromising it reaches all of them at once. That, and its being specific to this environment, is why it is the focus here.
Warehouse management systems (WMS) are the software platforms that optimize warehouse processes from the entry of goods to their exit. They provide real-time visibility into inventory and orchestrate end-to-end fulfillment by integrating with many warehouse subsystems and pieces of automation equipment, such as conveyors, PLCs, robotics, scanners, and voice systems. A typical WMS includes modules for receiving, inventory control, order management, labor management, shipping, and reporting. Major vendors include SAP EWM, Oracle Warehouse Management, Manhattan Active, Blue Yonder, Infor, Körber/HighJump, Microsoft, and TECSYS.
To picture how central this is, think back to our unicorn. From the moment it was scanned at receiving, it stopped being only a physical box and became a record inside the WMS, a record the system then used to decide where it should be stored, when it should be picked, how it should be sorted, and where it should ship. Every device employed along the way — the PDT at receiving, the printer at putaway, the workstation at picking, the conveyor at sorting — was really just talking to the WMS. It is the thread that ties all eight stages together, which is exactly what makes it such a valuable target.
Because modern WMS platforms are usually web-based and heavily interconnected, they inherit common web and API vulnerabilities alongside flaws specific to the WMS itself. A recent example is CVE-2026-70774 in Oracle Warehouse Management, part of Oracle E-Business Suite. Oracle rated it High and described it as an easily exploitable authorization bypass flaw in which a low-privileged attacker with network access can compromise the application, affecting integrity and availability by allowing unauthorized creation, deletion, or modification of critical data.
A flaw like this may look like a typical web application issue, but its consequences in a warehouse are more serious, because the WMS is the system of record. If an attacker can modify its data, they can alter stock levels, bin locations, receiving records, picking tasks, or shipment information, silently corrupting the digital twin of the warehouse. And since the eight stages form a dependency chain, that corruption doesn’t stay contained: a falsified record at receiving is inherited by putaway, picking, and everything downstream. The physical box and its record diverge, and every later process trusts the record.
The most direct version of this attack comes from an authorization bypass flaw like the Oracle vulnerability above. Here the threat actor can be an external attacker who has reached the application over the network, or an insider such as a low-privileged operator or storekeeper who already holds a legitimate account.
The danger grows when such a flaw is chained with others. An external attacker who combines remote code execution with SQL injection could generate shipment orders to arbitrary addresses together with the documents needed for goods to physically leave the warehouse, fabricate waybills, acceptance or transfer records, and registers, while also exposing customer data such as names and delivery addresses. In effect, the goods are sent out with legitimate-looking paperwork.
This is why ordinary web vulnerabilities carry considerable weight in a WMS. For an external attacker or an insider probing beyond their assigned role, broken access control could open a path into another warehouse zone, allow manipulation of other workers’ tasks, permit unauthorized exception operations, or expose administrative APIs. A server-side request forgery flaw could force the WMS server to reach resources an external attacker cannot touch directly, and if the network is flat or only formally segmented, that server may in turn reach internal management panels, cloud service interfaces, or systems sitting close to the operational network, turning a single web bug into a bridge between IT and OT.
Attacks on a WMS don’t always target the application directly, though. As we saw while tracing the unicorn, the warehouse is full of connected devices that already hold trusted access. A handheld terminal carries authenticated WMS API access as a matter of course, and anyone who gains control of that terminal inherits it (by escaping the kiosk, for example). Depending on the device’s permissions, that might mean confirming picks that never occurred, falsifying receiving operations, adjusting stock counts, or marking orders complete — something warehouse employees themselves could use to avoid their assigned duties.
Where the WMS is connected to control and automation systems, access to the WMS can extend into physical operations, including conveyors, sorting systems, lifts, doors, and pumps.
PLCs translate WMS instructions and sensor inputs into physical actions. Compromising a system that controls them therefore creates an opportunity to manipulate those actions. The impact can be as simple as a sorting diverter sending boxes down the wrong route, a conveyor repeatedly entering an error state and creating a backlog, or a falsified sensor value causing a line to be treated as clear when it is full.
However the WMS is compromised, the business consequence is the same: inventory corruption, unauthorized shipments, and operational disruption, each reaching the customer in the end as a late or missing delivery, the wrong item in the box, or a personal address exposed to a stranger.
We followed our unicorn through web applications and databases, radio channels and PDTs, printers, conveyors, robots, and the loading zone. Because warehouse environments combine IT, OT, wireless, physical, and human components, security controls should be implemented as a single, coordinated security architecture rather than as isolated measures. Network segmentation should be a foundational control, with clear separation between corporate IT, warehouse OT, operational devices, management networks, and externally connected systems. Segmentation should also extend across facilities and customer environments so that each site and customer integration is isolated from the others, limiting the potential impact of a compromise. A zero-trust approach should reinforce this architecture by requiring explicit authentication and authorization for connections between systems, regardless of their network location, and by applying least-privilege access wherever possible. Regular penetration testing of the corporate IT environment, including domain services, identity infrastructure, and endpoints, should be incorporated into the warehouse security program.
Physical and network access controls should be implemented together. Access to warehouse facilities, network rooms, communication cabinets, and operational areas should be restricted according to business need, with appropriate authentication, monitoring, and access logging. Network infrastructure should be hardened through controls such as switch port security, MAC-based access restrictions where appropriate, and IEEE 802.1X with EAP-TLS for strong device and user authentication. Network equipment should also be securely configured and regularly reviewed to ensure that unused interfaces, insecure management protocols, default credentials, and unnecessary services are disabled.
Web applications and their supporting infrastructure should be governed through a secure software development lifecycle (SSDLC), including security requirements, secure coding practices, code and dependency review, and application security assessment before deployment and throughout the application’s lifecycle. Third-party applications and components should be continuously monitored and updated, with a defined vulnerability and patch management process. This should explicitly include warehouse management systems (WMS), warehouse portals, APIs, and other applications that provide operational or customer-facing functionality.
Wireless technologies should be subject to dedicated security assessment and engineering. Radio-frequency surveys should be conducted to establish expected coverage, identify unauthorized or unexpected signals, validate the availability and protection of operational communications, and determine whether wireless deployments create unintended access paths into warehouse networks. Wireless security should be assessed not only from a connectivity perspective but also in terms of authentication, encryption, device provisioning, protocol exposure, interference resilience, and the potential impact of unauthorized devices.
Warehouse operational devices should be incorporated into the organization’s endpoint security and vulnerability management processes. Devices should receive security updates and firmware patches on a defined schedule, with accelerated remediation for critical vulnerabilities. Security assessments should be performed periodically to identify weaknesses in device configuration, applications, exposed services, and authentication mechanisms.
Personnel security should form an additional layer of the warehouse security architecture. In most industrial and ICS environments, getting inside means clearing background checks, holding relevant qualifications, or proving prior experience; the hiring process itself acts as a filter. A warehouse is different. Roles in sorting, packing, and courier work are hired quickly and in volume, and they typically require no prior expertise, certification, or specialized background at all. An ordinary person can walk in off the street, be hired within days, and immediately hold legitimate physical access to the facility. Therefore, a more restrictive and risk-based approach to hiring should be considered, particularly for roles that provide access to sensitive areas, operational systems, network infrastructure, or customer assets.
Ultimately, all of these measures serve the same objective: protecting the continuity and reliability of the warehouse as a whole. This matters because real warehouses handle food, medicines, machine parts, and the millions of ordinary things people and companies depend on every day. Protecting their journey means seeing the warehouse as it has actually become: a single digital system with doors, motors, shelving, and radio signals, and with entirely real consequences when it fails. And if the security program works, the customer never knows any of this happened. The unicorn simply arrives at the scheduled time.
If your warehouse operations rely on interconnected systems like these, we can help identify the gaps before an adversary does. Reach out to discuss a security assessment.
