Solar has released Solar Dozor 8.4 — an update to its DLP system that controls the transfer of data to public neural networks, Russian corporate messengers, and personnel documents. The practical change for administrators is that rules can be applied to specific channels and file types, rather than reducing policy to a general ban on workplace tools.
The new version includes monitoring and blocking features, a template for recognizing the first page of a military ID, and changes to workstation agents. Some capabilities depend on configured security rules: the mere presence of a feature does not mean every transfer will automatically be stopped.
Prompts and files before they are sent to a neural network
Solar Dozor analyzes prompts and files on the workstation before they are sent to public AI services. In its examples, the vendor names ChatGPT, Gemini, Claude, DeepSeek, and Perplexity, and also notes support for other services. The analysis covers both foreign and Russian neural networks.
Another change is that the system stores the history of AI conversations. When reviewing an event, information security specialists can see the context of a request, not just the isolated fragment that triggered a rule. This helps clarify exactly what data was transferred and what needs protection; the source does not specify how long the history is retained.
The administrator can configure one of two modes. Monitoring records events and generates reports, but does not interfere with employees’ work. In blocking mode, the system stops the transfer of confidential data if it violates configured policies, and displays a warning on the workstation.
The distinction matters for deployment: monitoring makes it possible to first build a picture of how services are being used, while blocking directly changes how the channel behaves. Still, the specific outcome will be determined by the organization’s rules. DLP without a configured policy is a bit like a security guard with a logbook who hasn’t yet been told whom not to let in.
According to Solar’s study for the first half of 2026, sensitive information is contained in almost 40% of requests to public AI services. In its breakdown of this data, the company classifies 41% as source code and configurations, 30% as personal, financial, and other sensitive information, 18% as intellectual property, and 11% as passwords, tokens, and API keys. These are the study results reported by the vendor, not a figure the source describes as universal for all companies.
Based on recorded events, the system generates reports on information categories and work scenarios involving its transfer to AI services. The infosec team can use these data to adjust policies and train employees using examples from everyday tasks. This approach does not replace the rules, but it makes it possible to rely on recorded usage rather than just assumptions about it.
Messengers and personnel files
Another focus of the update is Russian corporate messengers. In desktop and web versions on Windows, Linux, and macOS workstations, Solar Dozor can analyze messages and files or block the transfer of confidential information if it conflicts with security policies. The material does not list the specific messengers supported.
Events from these channels can be used to review information-sharing rules and prepare training examples for users. The source does not provide specific integration settings, deployment requirements, or a list of the message types that are analyzed. So before deployment, administrators should check the available capabilities against the configuration of their own environment.
For personnel documents, version 8.4 adds a ready-made graphical template that the built-in neural network uses to recognize images of the first page of military IDs. The infosec specialist needs to add the template to a policy rule; one example of its use is restricting the sending of copies by email. So the change concerns specifically identifying this image and configuring a rule, not automatically banning all personnel files.
The vendor describes the mechanism as a way to reduce the risk of accidentally transferring personal data through channels not permitted by the organization’s rules. The condition remains important: restrictions will work according to how the specialist configures the policy. The source also does not claim that the template recognizes all pages of a military ID or any other documents.
Changes to agents, servers, and the interface
On macOS, the agent has gained a self-protection mechanism. A PIN generated separately for each workstation is required to remove its components, stop services, or change settings. This adds control specifically over the listed actions involving the agent; the material does not describe detailed requirements for issuing or storing the PIN.
On Linux, the system now distinguishes between local and remote sessions belonging to the same user, so infosec specialists can identify the source of an event during an investigation. For Windows, the vendor says it has expanded monitoring of user activity and large-file transfers. The detailed parameters of this monitoring are not provided, so there is no basis for attributing specific additional logs or thresholds to it.
To improve performance, Solar Dozor 8.4 adds load distribution between servers and asynchronous execution of long-running operations in the background. The capabilities for analyzing archive contents have also been expanded. The description does not specify exactly how throughput, the list of formats, or processing time change, so the release makes no numerical promises about speed.
The web interface now lets users filter events by the author of the latest status change and select departments based on the organizational structure. Filters by department and job title have been added to the UBA module. When approving outgoing mail, it is now possible to allow several emails to be sent at once — a small change, but one that means less manual approval of each message individually.
Solar Dozor is positioned as a DLP system for protecting corporate data and analyzing leakage risks. According to the vendor, the product is listed in the Russian software registry and certified by the FSTEC of Russia at the fourth trust level. Update 8.4 combines new rules for AI, messengers, and a specific type of personnel document with agent and interface improvements; the actual behavior of the controls depends on the policies configured by the organization.









