Kaspersky Launches Malware Detection Feature for AI Agents
IT DAILY · · 1 views
✦ AI Summary
Kaspersky announced on the 18th a feature that helps verify security risks before adopting AI components.
The feature scans AI agents, models, and downloadable agent skills for malware and backdoors.
Kaspersky said it has identified more than 15,000 samples of malware disguising themselves as agentic AI software so far this year.
As AI agents increasingly show signs of a new supply-chain segmentation pattern that relies on external frameworks, APIs, and plugins, the importance of verifying the security of AI components themselves is rising. An "agent skill" is an extension designed for a specific task.
Kaspersky said on the 18th that it has introduced a feature that extends security for the AI sector from its existing SW supply-chain security area, helping enterprises identify security risks before adopting AI components in their environments. The announcement covers malware and backdoor scanning for "agent skills."
This malware and backdoor scanning feature for AI agents operates beneath Kaspersky's existing endpoint and threat intelligence products. The scanning targets AI agents, models, and downloadable agent skills.
For the "OpenClaw" skill, scanning is performed at the point when a user downloads it. The "OpenClaw" skill also supports scanning at the time of actual implementation.
Through this feature, Kaspersky said it has enabled companies to examine related security risks before adoption in environments where AI components are expanding in supply-chain form.
Kaspersky said that as a result of testing agentic AI skills provided through public libraries, vulnerabilities related to those skills continue to be discovered.
Kaspersky also said it has identified more than 15,000 samples of malware disguising themselves as agentic AI software so far this year. Earlier, the company detected more than 92,000 attacks between 2026 and early May involving malware and potentially unwanted application campaigns impersonating AI agents and AI services worldwide, with fake ChatGPT applications accounting for the largest share of detections at 49%, while Claude and Gemini each accounted for 18%.
Adrian Hia, president of Kaspersky for APAC, said AI agents are forming a new supply-chain layer and that agents depend heavily on external frameworks, APIs, and plugins. He added that a compromised upstream supply-chain component can spread in a chain to related systems.
OpenClaw is a term referring to an open-source AI agent that integrates external tools and services.
This service can expand its functionality by adding separate "skills." The developer is Peter Steinberger of Austria.
OpenClaw was first unveiled in November 2025. After its release, it became a global topic, including in South Korea, around February this year.
Source: IT DAILY · Kim Ho-jun
Original: https://www.itdaily.kr/news/articleView.html?idxno=241037
References
This article was produced with the help of an automated content generation algorithm.
Source: IT DAILY
View originalThis article was summarized and organized by BizCrush based on the original article from IT DAILY. For exact quotations and full details, please refer to the original article.