Reports this week say Cyera is close to closing a round of roughly $800 million at a valuation near $20 billion, with Cyberstarts expected to lead and contribute about $300 million. Cyera has not yet confirmed the round, but the reported size tells us something about where the market is going. Investors are betting that governing AI agents will become a core security category.
At Virtru, we agree. The harder question is where that governance should be enforced.
Cyera's answer is Agent Guardian and Cyera Endpoint. Agent Guardian discovers AI agents, including ones nobody approved, maps the data and tools each one can reach, and operates inline to intercept tool calls, block unauthorized data transfers, and quarantine compromised agents. Cyera Endpoint extends that to local tools such as Claude Code and Cursor, which run outside network monitoring, and protects sensitive files on the device before they are accessed or leave it. This is a legitimate approach to a real problem. Most enterprises have far more agents running than they know about, and finding them is the first step.
The question is what the control is attached to. Every enforcement point in this model sits around the data, rather than on it: a platform API, an AI gateway, a browser extension, a framework hook, or a remote scan of an employee's device. The agent is permitted to reach the data, and the control tries to stop it from doing the wrong thing with what it reached. That works while the proxy is in the path. Once the data is loaded into a model's context, handed to a partner, or copied to a personal AI account, the protection depends on whether the proxy still sees the traffic.
This is not unique to Cyera. The large majority of vendors in this market, from identity platforms and AI gateways to runtime monitoring and posture tools, govern agents at the application or agent layer. They decide what an agent may call and watch what it does while it works. We think this approach is incomplete, for three reasons.
Virtru takes a different path. The Virtru Data Harbor Platform is built on the Trusted Data Format (TDF), an open standard that cryptographically binds attributes and policy to each data object. When an agent asks to access data, two checks run. The customer's own policy engine decides whether the request is allowed, and the agent's entitlements are tested against that object's attributes. The decryption key is released only if both pass. If either fails, the object stays encrypted, wherever it is. No key, no plaintext.
That design changes things:
To our knowledge, Virtru is the only vendor using TDF to cryptographically control and enforce AI agent access to the data itself. Public sector and commercial enterprise customers have been utilizing this approach for years, and the feedback has been consistent.
One global financial institution assessed a long list of AI security vendors. Each could detect agent activity or decide entitlements, but none enforced policy on the data once an agent had legitimate access. In six weeks, the bank and Virtru built a working prototype on the Data Harbor Platform. It ran self-managed and air-gapped, with keys and policy staying inside the bank, and agents could access only the data they were entitled to. In the bank's words: "We can prove what data was accessed and what was refused."
Cyera and Virtru are not pursuing the same problem in the same place, and enterprises may need both. Runtime controls answer whether an agent should be doing something right now. Object-level enforcement answers whether this agent may access a piece of data, for this purpose, under this policy, and it keeps answering that question as the data travels.. As agents begin to handle sensitive data across more organizations, we expect the second question to become increasingly critical.
Join us on Tuesday, November 17, for the launch of the Virtru Data Harbor Platform. We will show how enterprises and agencies enforce access policy on the data itself, one object and one request at a time, and hear from the partners who built on the platform.