WhiteMatter raises $20 million to put policy controls around enterprise AI agents

guthlabs.newsroom | AI Security | November 10, 2026
WhiteMatter, the AI-security company previously known as QuilrAI, announced a $20 million Series A on November 10. Blumberg Capital led the round, with existing investors Crew Capital and Sprout & Oak participating. The company plans to spend the funding on research, product development, commercial expansion and executive hiring. WhiteMatter funding announcement
Its proposition is to place governance around agents before deployment and around the actions they take afterward. WhiteMatter says it can assign a dedicated Guardian agent to an enterprise agent, adding more Guardians as the customer’s agent population expands. The platform is intended to enforce policy across models, tools and integrations, including Model Context Protocol servers. WhiteMatter funding announcement
The company names Sentara Health, Alter Domus, New American Funding and Concentrix among customers. It reports fivefold revenue growth over six months but does not disclose the underlying revenue figure in the announcement. Customer examples and operating statistics in the release are company-supplied evidence, rather than independent tests of the platform’s security effectiveness. WhiteMatter funding announcement
From employee security to agent governance
The new financing follows an earlier business focused on using agents to reduce enterprise security risks. In an April 2025 investment post, Crew Capital described Quilr agents handling data protection, identity risks, employee guidance, workflow automation and device issues. The investor said the product connected with collaboration tools, browsers and other IT systems, and that it had led Quilr’s seed round. That earlier announcement provides context for the company’s development; it is not a new funding event this week. Crew Capital’s original investment post
WhiteMatter’s current positioning extends that approach to the growing number of software agents operating inside organizations. The underlying problem changes when a system can select tools and carry out a sequence of actions: a policy must govern what happens next, rather than simply classify a finished response.
Consider an agent authorized to summarize a customer record but not send it outside the company. A useful control needs to understand the distinction between reading, summarizing and transmitting. It also needs enough context to identify the recipient and data involved. Checking only the model’s final text would miss some consequential activity occurring through tools along the way.
The supervisor needs testing too
Using an AI agent to supervise another does not remove the need for conventional security boundaries. The supervisory system needs a defined scope, reliable enforcement points and a clear response when it cannot make a decision. Buyers should establish whether an unavailable policy service blocks an action, permits it or transfers it to a person.
Testing should include unauthorized tool calls, misleading instructions in retrieved documents, requests that cross data boundaries and actions that produce unexpected side effects. Equally important is the rate at which legitimate work gets blocked. A control that creates constant false alarms can become an operational burden even if it detects genuine threats.
The financing gives WhiteMatter resources to expand a product category built around those questions. Its next meaningful proof will be detailed evidence of what its controls cover, how they fail and how well they work in customers’ actual agent environments. A larger deployment footprint can demonstrate adoption; dependable enforcement requires a separate standard of evidence.
Sources and citations
Revision history
Published .