Eset expands its Protect platform to secure chatbots, AI agents, and malicious skills in the enterprise

Generative AI tools are becoming commonplace in daily professional use, often operating without IT supervision across many organizations. Eset is enhancing its Protect platform with new capabilities that include monitoring user-AI interactions, protecting agent software supply chains, and detecting malicious skills prior to deployment. A free, no-registration-required skill analysis tool complements these offerings.

The unsupervised use of public cloud chatbots is contributing to the phenomenon of “ghost AI,” where sensitive data—including internal documents, API keys, and credentials—transfers to third-party services beyond IT oversight. This can lead to compliance breaches, particularly regarding GDPR, as well as exposure to manipulated content or deliberately misleading responses.

Eset is integrating these new protections into Eset Protect, its centralized administration console. The solution relies on a secure browser deployed on endpoints that intercepts real-time interactions between employees and AI tools, logging exchanges locally and sending the data back to the console for analysis.

Real-Time Monitoring of Interactions

The secure browser analyzes each interaction with AI tools by scrutinizing user-submitted prompts and the responses received. This approach enables real-time detection of malicious URLs embedded in the responses from large language models (LLMs), prompt injection attempts, suspicious scripts, and sensitive data entries. Organizations can configure rules to block or monitor these actions as per their internal security policies.

Centralized activity logs in Eset Protect are available for Security Operations Center (SOC) teams to investigate and enforce internal policies. For entities subject to GDPR or DORA, this system provides traceability elements necessary to demonstrate that sensitive data flows remain within the organization’s control, even when employees use third-party AI tools in their daily activities.

Agentic AI and Software Supply Chains: New Attack Vectors

The rise of agentic AI introduces attack vectors distinct from those associated with conversational chatbots. Threats now target agent orchestration frameworks and third-party libraries incorporated within autonomous pipelines. LiteLLM, widely utilized in multi-agent architectures, has been identified as a target for trojanization. OpenClaw-type agents, capable of interacting with file systems or executing code under minimal human supervision, represent another vector for exfiltration or takeover.

Eset states its mechanisms already cover components distributed through standard repositories, with ongoing research aimed at anticipating the evolution of these attacks. Their participation in the Agentic AI Foundation extends this approach, as the company collaborates with other industry players to define secure communication standards for AI agents and best interoperability practices.

Eset AI Skills Checker: A Free Behavioral Analysis Tool

Eset is offering a free skill assessment tool on its website, requiring no account or subscription. The tool supports skills from major repositories (ClawHub, playbooks.com, skills.sh, GitHub, GitLab) and provides an evaluation in three levels (safe, dangerous, or suspect) by examining the raw content of the skill file, extracting and verifying each referenced external URL, and tracing the entire download chain.

A distinctive feature of the tool is its behavioral simulation capability. It executes the skill in a cloud-based sandbox dedicated to AI agents, allowing for the identification of malicious behaviors that only emerge during prolonged interactions, which static analyses would not detect. This multi-layered analysis logic is also integrated into Eset’s Endpoint solutions for organizations already utilizing the platform.

Eset’s announcement reflects a structural expansion of cybersecurity within enterprises. “For over thirty years, Eset has provided security solutions based on AI and machine learning, helping organizations counter the increasing use of AI by cybercriminals to conduct sophisticated and automated attacks,” emphasized Juraj Jánošík, Eset’s Director of Artificial Intelligence. For CISOs, the challenge now transcends mere tool oversight, encompassing governance over the entire execution chain of autonomous agents deployed within the organization.

Scroll to Top