Exabeam is expanding Agent Behavior Analytics with support for ChatGPT, Microsoft Copilot, and Google Gemini. Five new features monitor AI agent behavior for anomalies, detect prompt injection, and verify access rights. The expansion also includes coverage for the most significant agentic AI risks and offers enhancements to the New-Scale and LogRhythm SIEM platforms.
These additions are designed to close the security gap created by virtually invisible AI agents. A lack of visibility into employees’ use of AI—or “shadow AI”—poses a variety of risks. These risks stem from the fact that agents, just like their human “colleagues,” can abuse privileges, leading to data breaches or infiltration by cyberattackers.
AI agents are, in fact, being deployed more and more actively as digital employees. They authenticate, access systems, and independently execute business processes. A savvy cyberattacker can disrupt these processes by, for example, misleading the agent with false information. Even unintentionally, an AI agent can exhibit undesirable behavior if it has too much freedom of action. Exabeam aims to put a stop to this.
Five new features
Exabeam is therefore launching five new features that, according to the company, collectively cover the entire agentic attack surface. AI behavior baselining first builds dynamic behavior profiles for users and their AI agents, and tracks patterns in request volumes, token usage, tool calls, and outbound traffic. Deviations are immediately flagged to security teams, ensuring human intervention occurs as quickly as possible.
In addition, Exabeam introduces prompt and model abuse detection, designed to detect prompt injection, model manipulation, and the exploitation of tools. This is accompanied by a detection library that is five times larger than the previous version. Identity and privilege monitoring and agent lifecycle monitoring are also new: the former monitors what agents are permitted to do, while the latter provides insight into the full lifecycle of agents.
The fifth feature is coverage for the OWASP Top 10 for Agentic AI. This measures agent behavior against the framework of the Open Web Application Security Project, a framework that was previously lacking for this category of threats. While this coverage is not explicitly comprehensive, it does address the most common and high-risk issues.
Growing Threat
The expansion is a next step from what Exabeam released in January, when the company introduced functionality for AI agents into its New-Scale platform. At that time, the focus was still on behavioral analysis as the core mechanism; now, support for ChatGPT, Copilot, and Gemini has been added.
Previous research shows that the need is urgent. Studies revealed that AI-driven insider threats already outnumber external attacks worldwide. Given that about 40 percent of business applications will integrate AI agents this year, compared to less than 5 percent in 2025, now is the time to prevent dangerous adoption.
The new features are accompanied by other enhancements to the Exabeam New-Scale and LogRhythm Platforms. These are designed to support the daily work of administrators and security analysts, with the goal of reducing alert fatigue and enabling faster detection.