Cyberhaven has launched a security product designed to show companies what AI agents do with their data and stop sensitive information leaving. Its pitch is straightforward: agents can act quickly with the access granted to the people they serve, so organisations need visibility into what they touch and where it goes.
Watch Desk analysis
What happened
Cyberhaven calls the new module Agentic Data Security, or ADS. The company says it can discover AI tools and agents across endpoints, cloud services and browsers, including unsanctioned apps; record prompts, tool calls, file reads and responses; link activity to the person or system behind it; and apply existing data policies to block risky transfers.
The launch was reported by Help Net Security on 5 November. Cyberhaven says ADS is part of its Flow platform and works across endpoints, browsers and the cloud. The company also says its Labs team found endpoint-based agentic AI app adoption doubled year on year, reaching 60% in May 2026. That is Cyberhaven’s own measurement, not an independently established industry-wide figure.
Why it matters
An agent may be doing exactly what its user asked while still reading or sharing information that user should not have exposed. Cyberhaven’s proposed answer is to follow the data through an agent’s workflow, identify who authorised the action and apply existing controls before sensitive material leaves.
That makes the product relevant to a practical problem for companies moving agents beyond pilots: how to allow useful automation without giving every agent a free pass to company files. The launch describes a set of security controls, not proof that they catch every risky action or prevent every data leak.
Our read
The useful idea here is to secure the data’s journey, rather than simply keep a list of which AI tools are approved. If the promised visibility and enforcement work across the services companies actually use, that could give security teams a more workable way to let agents in without handing them the keys to the filing cabinet.
The test will be whether those controls are clear and reliable in practice, especially when agents use multiple tools or handle information across different services. A tidy product description is not the same thing as a tidy incident log.
What to watch
- Whether customers can inspect and enforce policies across agents and services they already use.
- How Cyberhaven defines and measures the reported 60% adoption figure.
- Whether the company publishes evidence of how ADS identifies and blocks risky data movement.
Discussion spark: Should companies let AI agents use the same data permissions as the employees they serve, or require agents to have narrower access by default?
Sources and evidence
- Source update (5 November 2026, 15:04 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.