Cyberhaven introduces controls for tracking AI agents’ use of company data
Cyberhaven has introduced Agentic Data Security (ADS), a product module intended to help organizations monitor and control how AI agents interact with company data. Announced on November 5, 2026, the offering is aimed at security, insider-risk and AI-governance teams as businesses move agents from trials into operational workflows.
The launch addresses a security challenge that differs from conventional employee activity: an agent may use the permissions of the person or system directing it, then read files, query databases and pass information between tools as part of a single workflow. That can make it difficult for teams to determine what information an agent accessed, who authorized its activity and where the information went afterward.
Cyberhaven describes ADS as a way to apply data lineage and existing security policies to agent activity. The company says the module can discover AI tools and agents across endpoints, browsers and cloud environments, including unsanctioned applications, and assign risk scores. It also says the product records prompts, tool calls, file reads, reasoning traces, responses and related data so security teams can inspect agent workflows.
Visibility and enforcement across agent workflows
The product’s controls are organized around discovery, activity monitoring, attribution and intervention. Cyberhaven says ADS links an agent’s actions to the person or system behind them, presenting human and agent activity together. Its stated enforcement function applies policies already used for people to agent actions, with the aim of containing sensitive information before it leaves the organization.
Cyberhaven says the module operates across endpoints, browsers and cloud environments and uses the same policy engine as its broader platform. The company positions this as a way to add agent controls without introducing another console or a separate policy set. The announcement does not specify deployment requirements, supported agent frameworks, pricing, or a detailed list of integrations, so organizations evaluating the product will need to validate those details with the vendor.
ADS is the latest module in Cyberhaven Flow, which the company describes as a unified data security platform spanning data loss prevention (DLP), insider risk management and data security posture management. Cyberhaven says its platform uses data lineage, AI and automated protection, and that the new module extends the same approach to both human and agent workflows. The company also connects the launch to a strategy it introduced in July, combining data lineage, identity and behavior to protect data as it moves through workflows.
Adoption creates a new data-governance challenge
Cyberhaven Labs reported that endpoint-based adoption of agentic AI applications doubled year over year, reaching 60% in May 2026. That figure is the company’s finding; the announcement does not provide the study’s methodology or define the population measured. Even without treating it as representative of every organization, the reported increase reflects why security teams are seeking ways to identify agents operating beyond formal pilots.
Cyberhaven CEO Nishant Doshi said agents can work with access granted to trusted employees, and warned that an incident could involve an agent using data it was not meant to access while carrying out an assigned task. His point underscores a distinction important to defenders: legitimate credentials and an apparently authorized workflow do not, by themselves, establish that a particular data use is appropriate.
Lawrence Pingree, head of research at SACR, said effective governance requires understanding which data was accessed, by whom, through what workflow and whether the movement was authorized. Datavant CISO Dan Walsh likewise emphasized the speed and scale at which agents can move data when operating with trusted access. These comments reflect the broader operational question ADS is designed to address: connecting an action to its context, rather than relying solely on the identity of the account or application involved.
For enterprise security teams, agent discovery and data-level controls can help bring unsanctioned or overlooked use into existing governance processes. The usefulness of such controls will depend on the completeness of activity visibility, the accuracy of attribution and how enforcement policies distinguish legitimate work from risky data movement. Teams should also establish who can approve agent access and review recorded activity, particularly where prompts, tool calls and traces may contain sensitive business information. Cyberhaven’s announcement describes the intended capabilities, but organizations should assess their coverage and policy behavior in their own environments before relying on them to prevent data loss.
