Kaspersky products can detect and block malicious AI agent activity on endpoints
The claim
Kaspersky’s security suite can detect, control, and mitigate threats from AI agents on corporate endpoints.
Published as: “Deploying AI agents securely: how Kaspersky products mitigate AI threats”
So what
Organizations can adopt AI agents while reducing risk by leveraging Kaspersky’s layered detection and control capabilities.
The five-minute version
The post describes how Kaspersky integrates AI-agent profiles, specialized EDR rules, Application and Web Control categories, and a Scan Engine that inspects AI skills and LLM files. It cites telemetry showing the popularity of AI tools and reports over 2,500 AI-related vulnerabilities published since 2025. The evidence consists of internal telemetry, a case study of the Armored Likho campaign, and rule definitions, but no independent third-party validation.
What they found 5
- More than 2,500 AI-related vulnerabilities have been published between 2025 and 2026.
“From 2025 to date, more than 2500 vulnerabilities have been published:”
- Anthropic and OpenAI tools are the most popular AI agents among Kaspersky users in September 2026.
“Tools from Anthropic and OpenAI take the top spots in the September 2026 ranking of AI agents by popularity among Kaspersky users.”
- Kaspersky observed AI-generated loader and dropper samples in the Armored Likho campaign.
“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).”
- Dedicated EDR rules generate alerts when AI agents access credential files.
“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.”
- A rule detects AI agents running in auto-approval (YOLO) mode or with sandbox disabled.
“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.”
What undercuts it the paper’s own results
- unsafe actions by an agent do not always indicate an attack.
- The mere fact that an agent was launched does not reveal whether it is being used in an attack.
Stated limits 2
- These settings do not indicate an attack on their own, but they let the agent perform dangerous actions without user confirmation.
- unsafe actions by an agent do not always indicate an attack.
What’s new here 4
- AI agent profiles added to the Kaspersky Threat Intelligence Portal.
- Specialized EDR detection rules for AI agents (e.g., ai_agent_*_access_to_sensitive_file).
- Application Control and Web Control categories for allowing or blocking AI tools and AI-as-a-Service sites.
- Kaspersky Scan Engine now scans LLM model files and AI agent skills for malicious code.
Artifacts & methods 8
- Kaspersky Endpoint Security / Kaspersky Nextsystem
- Kaspersky Endpoint Detection and Response (KEDR)system
- Application Controlprimitive
- Web Controlprimitive
- Kaspersky Threat Intelligence Portaldataset
- Hunt Hub detection rule repositorydataset
- Kaspersky Scan Enginesystem
- ai_agent_* detection rulesprimitive