How to protect a company from attacks through AI agents | Kaspersky official blog
Source Kaspersky Security Blog
Description
This analysis discusses the emerging security risks associated with AI agents used in corporate environments. Attackers exploit AI assistants by sending them malicious plain text prompts to perform unauthorized actions such as data exfiltration. Risks include unapproved AI tools (Shadow AI), fake AI software installers delivering malware, excessive permissions granted to AI agents, malicious MCP servers connected to AI applications, and malware that directly queries large language models (LLMs) for malicious code. Traditional security tools like application control, web filtering, endpoint detection and response (EDR), and threat intelligence can mitigate these risks. Monitoring AI agent permissions, instruction sources, and behavior is critical to preventing abuse. The article emphasizes the importance of maintaining approved AI tool registries, restricting unauthorized AI service access, and vetting connected components.