ESET has equipped its ESET Protect platform with an AI security module designed to monitor the components and actions of autonomous agents within enterprises. This module leverages the rapid proliferation of these agents, some of which have been flagged as malicious by ESET’s researchers.
Launched on September 30, the new AI security module for ESET Protect is tasked with analyzing the components of autonomous agents and monitoring their activities. According to ESET, this module allows businesses to adopt these agents without needing to multiply tools, especially in sectors where the use of AI is advancing faster than appropriate regulatory frameworks can be established.
This rollout completes a series of announcements ESET made earlier in the year. In April, the company revealed measures against AI supply chain attacks and introduced a free tool, the AI Skills Checker, for assessing agent competencies. By the end of August, during Black Hat, ESET elaborated on the capabilities aimed at agents that were officially launched on September 30.
Rapid Proliferation of Agent Competencies
On March 20, ESET researchers reported over 60,000 agent skills, with 72% classified as new. A skill consists of a package of instructions and scripts that an agent loads to perform a function. By May 19, this figure had surged to nearly 800,000 unique skills vetted since March—thirteen times the initial total—with about 25,000 deemed suspicious and over 3,000 flagged as malicious. In its first-half threat report, ESET cited 900,000 analyzed skills, maintaining the same numbers for suspicious and explicitly malicious categories.
A consistent theme throughout these findings is the rapid expansion of agents and their competencies. The flagged skills are associated with behaviors such as executing commands or downloading third-party tools, described in the report without being classified as attacks. Each new agent added effectively enlarges an enterprise’s supply chain, underscoring the necessity for component monitoring.
Monitoring Components and Agent Behavior
The new module comprises two key capabilities. AI security for agents examines the components used by these agents—namely their skills, extensions, and dependencies. It blocks malicious or compromised elements, such as external addresses, from reaching enterprise resources. ESET sees this as a solution to the risk posed by AI supply chains without requiring manual checks on every component. Behavioral monitoring tracks the actions of these agents, detects unusual access, and alerts teams about activities outside the expected perimeter, allowing them to intervene before an incident occurs. ESET notes that these capabilities are integrated into its AI technologies and managed detection and response services.
ESET pairs this module with two advancements in AI-assisted detection. ESET LiveCortex correlates large volumes of events to distinguish significant signals from background noise, while ESET AI Advisor has evolved into ESET LiveAI, providing incident explanations and remediation measures within a controlled framework. All of this is hosted on the same platform, minimizing the need for multiple tools, according to ESET.
Essential Oversight Before Deploying Agents
ESET references guidance from the CERT-FR bulletin dated April 13 regarding automated agent products for workplaces. The agency advises against deploying autonomous assistants in production environments until they are stable and suggests isolating their use to test environments devoid of sensitive data. For any subsequent deployment, explicit approval rules, human validation of system commands, compartmentalized environments, and authorized communication channels are required. This recommendation has been duly noted.
The new module meets part of these requirements by analyzing components and monitoring actions, while decisions on compartmentalization, human validation, and authorization rules remain with the enterprise itself. This is promising for CIOs preparing for controlled tests, as the tool integrates seamlessly with an existing security console. The module’s recent debut means no independent testing results are available yet. A referenced test by AV-Comparatives from July 1 assessed ESET Protect Elite Cloud’s detection capabilities, noting a success rate of 11 out of 14 attack steps detected, not accounting for agent security. Consequently, the policy regarding agent permissions must be determined before they are introduced to users.
Expanded Platform with a €40 Million Investment
On May 19, during ESET World in Berlin, the company announced an additional investment of €40 million slated for 2026. This funding aims to develop security-oriented core AI models, a native AI security architecture including ESET Secure AI Relay, and technologies for AI-assisted security operation centers. Michal Jankech, ESET’s Vice President for Enterprise and SMB/MSP, highlights the priority for organizations lacking specialized teams—particularly small to medium enterprises—will focus on controlling access, data, and actions executed by these tools. ESET plans to expand its research and development team to 1,000 researchers and engineers over the next three years.
This follows a trajectory from RSA Conference to September 30, signaling a movement by the endpoint security provider towards offering small and medium-sized enterprises and their service providers a unified platform for agent control. For CIOs, this centralization simplifies operations and necessitates verification of which capabilities are included under which licensing for ESET Protect within France.



