Cyberhaven launches Agentic Data Security to track and control AI data use

Cyberhaven has launched Agentic Data Security (ADS) to help security teams manage the risks associated with using AI agents on company data while maintaining control. This solution provides visibility into what AI agents do with company data, who they are acting for, and where the data moves, along with the ability to stop sensitive data from leaving. The introduction of AI agents has created new insider risk vectors because agents can operate at machine speed with the access of trusted employees across various workflows.
The Agentic Data Security module is part of Cyberhaven Flow, a unified platform integrating data lineage, AI, and autonomous protection to deliver DLP, IRM, and DSPM for both humans and agents. The platform focuses on understanding the entire lifecycle of data by tracing it through data lineage, identity, and behavior across workflows involving agents.
The system offers four key controls: finding all agents, seeing agent actions (prompts, reads, traces), knowing the origin of actions (who sent what), and stopping sensitive data from leaving by applying policies to agentic actions. The company posits that controlling the data agents touch, rather than just deploying the agents, is the defining factor for security in the agentic enterprise.

* Cyberhaven introduced Agentic Data Security (ADS).
* ADS aims to help CISOs, data security teams, insider risk teams, and AI governance leaders manage AI agents without losing data control.
* ADS shows what every AI agent does with company data, who it acts for, the data's next location, and stops sensitive data before leaving.
* Endpoint-based agentic AI app adoption doubled year over year, reaching 60% in May 2026.
* Agentic Data Security is a module in Cyberhaven Flow, a unified data security platform.
* The platform uses data lineage, AI, and autonomous protection to deliver DLP, IRM, and DSPM for humans and agents.
* ADS provides four controls: finding agents, seeing actions (prompts, calls, reads), knowing the sender, and stopping data before leaving.
* Cyberhaven claims that agentic enterprise security depends on controlling data agents touch, not just deployment.
* The system applies data lineage and a unified policy engine to agents without adding separate tools or policies.

The narrative centers on the tension between rapid technological adoption—the proliferation of AI agents with trusted access—and the lagging security controls necessary to govern this new operational space. The core implication is that the risk profile shifts from traditional perimeter security to the integrity of autonomous workflows. Agents, acting at machine speed with employee-level access, represent a novel class of insider risk where malicious or accidental exfiltration can occur through legitimate, documented steps.
The emphasis on data lineage as the foundation for agent governance suggests that visibility into the *process* of data movement is more critical than simply monitoring endpoints. The move from human oversight to agentic execution forces security concepts like Data Loss Prevention (DLP) and Insider Risk Management (IRM) to be extended seamlessly across non-human actors. This moves security strategy toward continuous, context-aware enforcement rather than static policy checks.
The concept of control over the data itself, rather than just the tools touching it, points toward a potential systemic shift in accountability. If policies can be applied directly to agent actions by inspecting their reasoning traces and tool calls, it suggests a future where security posture is intrinsically woven into the operational fabric of AI systems. The challenge lies in ensuring that this technical capability translates into enforceable, equitable governance that does not impede the innovation agents enable, and critically, that the lineage tracking itself remains trustworthy and unmanipulated by the very agents being monitored.
Bridge Questions: If data lineage becomes the primary mechanism for agent governance, what ethical or legal accountability framework must evolve to assign responsibility when an autonomous agent executes a flawed but authorized workflow? How can organizations ensure that the speed of agentic deployment does not outpace the necessary development of these unified policy engines across complex organizational hierarchies? What are the potential systemic risks if security measures become so deeply integrated into agent workflows that they risk becoming the new bottleneck for innovation?