Cyberhaven launches security tool to monitor AI agent data access

The rise of agentic risk

AI agents are shifting from experimental pilots to active production environments across various industries. These autonomous tools can perform complex tasks like reading files, querying databases, and sharing findings within a single workflow. Because they operate with the permissions of the people they serve, they can move data at machine speeds.

This rapid adoption creates a new category of insider risk for security leaders. Cyberhaven Labs reported that endpoint-based agentic AI app adoption reached 60% in May 2026, doubling its previous year-over-year growth. This means many companies are deploying tools faster than they can build the necessary controls to manage them.

The primary concern for Chief Information Security Officers (CISOs) is the lack of visibility into what these agents are doing. A single agent might perform an action that is technically correct based on its instructions but involves data it should never have touched. Cyberhaven aims to provide a way to say yes to these tools without losing control of the underlying information.

How Agentic Data Security works

Cyberhaven's new Agentic Data Security (ADS) module is part of a larger platform called Cyberhaven Flow. It uses a combination of data lineage, AI, and autonomous protection to manage data loss prevention (DLP) and insider risk management (IRM). Data lineage refers to the ability to trace the full lifecycle of information as it moves through various systems.

The platform works by documenting agent actions as content that can be inspected by security teams. It captures specific details such as prompts, tool calls, file reads, and reasoning traces. By treating these actions as documented data, the system allows for full transparency into the agent's decision-making process and the specific data it involves at every step.

The technology applies a unified policy engine to both humans and AI agents. This means that the rules a company already has in place for employees can be applied to the actions of an AI agent. The system identifies risky behavior and contains sensitive data before it leaves the corporate environment, rather than simply blocking the use of the AI tool itself.

Visibility for security teams

One of the core features of ADS is the ability to discover every AI tool and agent running across a company's infrastructure. This includes apps running on endpoints, within clouds, or in web browsers, even if they are unsanctioned applications. The platform then scores each of these tools based on the level of risk they pose to the organization.

Security teams can also see exactly who sent a specific request to an agent. By tying every action back to a person or system, the tool provides a unified view of both human and agentic workflows. This helps teams understand the chain of command and identify which users are interacting with which autonomous tools.

The platform also helps with data classification and flow analysis. Cyberhaven uses proprietary AI models to help customers understand how data is transformed as it moves through an AI workflow. This allows the system to protect data whether it is at rest in a database or in motion as it is being processed by an agent.

Impact on the enterprise

For companies in high-stakes industries like frontier AI labs or quant hedge funds, managing data integrity is critical. These organizations already rely on Cyberhaven to protect their most sensitive information. The new ADS module is designed to give these leaders the confidence to lean into the agentic future without watching their data walk out the door.

Industry analysts suggest that governing risk in the AI era requires a deep understanding of what data was touched and by whom. They note that Cyberhaven's focus on data lineage provides the necessary foundation for managing both people and AI agents in a single platform. This approach moves away from just securing the tools and focuses on securing the data itself.

The goal is to define the agentic enterprise by how well a company controls the data its agents touch. As agents move faster and operate at scale with trusted access, the math on data exfiltration changes. Organizations that can successfully implement these controls will be able to adopt AI more broadly while maintaining their security posture.