Israeli-founded Sentra announced a $150 million investment led by Capital One Ventures on November 4 to expand enterprise data security for AI agents. Co-founder and chief technology officer Ron Reiter described the financing as a Series C round. A representative of the investor will join Sentra's board.
The money will support product development and commercial expansion, but the announcement centres on a change in who consumes corporate information. Alongside employees and applications, autonomous agents read documents, consult knowledge stores and perform tasks. Broad permissions once granted to a service account can now give an agent access to sensitive information well beyond its assignment.
Funding the data layer agents can reach
Sentra named existing investors Bessemer Venture Partners, Zeev Ventures, Standard Investments, Munich Re Ventures and Key1 Capital. It said the investment followed annual growth exceeding 300%. The announcement does not identify the business metric behind that percentage, so it should not be treated as a defined revenue or customer-growth figure.
The company plans better visibility into agents, AI assistants and models and the sensitive information they encounter. It also intends to expand continuous least-privilege enforcement and rapid responses to unusual access. Commercial priorities include sales channels, partners, marketing and new regions; Reiter specifically mentions a deeper presence in North America and Europe.
Its stated approach starts with stored data, before an agent requests it. Sentra says it maps information across clouds, software services, data warehouses and on-premises file servers. That mapping addresses two separate questions: what is sensitive, and who can reach it. Answering only one leaves part of the exposure unexplained.
A document, for example, may be correctly classified but shared with too many people. Conversely, an apparently reasonable permissions list does not reveal whether a folder contains marketing material or customer records. Connecting content with identities is central to Sentra's proposition. It aims to reduce exposure before an agent starts work, without relying on the agent to recognise every problem.
Before an action, during an incident and afterwards
Sentra describes three stages of protection. Before an action, its system is intended to discover and classify sensitive information and map access for employees, service accounts and agents, including inherited permissions. The company also describes removing redundant, obsolete or high-risk data. The aim is to reduce what an agent could find in the first place.
During unusual activity, Sentra offers detection informed by data sensitivity, permission revocation and removal of risky sharing through automated remediation. High-risk issues go to data owners and existing workflows. These are vendor-described capabilities, rather than a promise that every unauthorised action will be blocked in every system or customer configuration.
After an incident, the question becomes not just what happened but what the agent could have reached. Sentra says it presents potential exposure with the datasets' business context. Potential access is not proof that information was actually read or transferred. An investigation must distinguish permissions that existed from actions recorded in logs.
Consider an agent authorised to summarise support requests. If it can also read a payroll folder, exposure exists before it produces an incorrect summary. Restricting permissions addresses its scope of action; checking its answer addresses its output. The two controls can complement each other, but measure different things and cannot replace one another.
Data stays in the customer's environment, Sentra says
Sentra emphasises analysing information where it resides, inside the customer's environment. Reiter's post says the scanning engine runs there and only metadata leaves. This is the supplier's description of its architecture. An organisation considering deployment still needs to establish which metadata is transmitted, who can access it and how long it is retained.
Reiter describes an evolution from cloud Data Security Posture Management, or DSPM, to a broader platform. Alongside discovery and classification, it covers data-access governance, loss prevention and treatment of unnecessary copies and exposed secrets. He cites small language models and business context for classification. Using a model does not itself establish classification quality in every organisation.
A practical assessment starts with familiar business examples. Does the system identify a sensitive document whose filename reveals nothing? Can it distinguish a working copy from a record that must be retained? Could revoking access interrupt a critical process? These are evaluation questions arising from the product category, not published findings across all Sentra customers.
The announcement also includes Lyft security chief Chaim Sanders describing the need to know what data agents can reach before switching them on. His account illustrates the problem Sentra aims to address. It does not replace measurements of detection, false alarms or response time in another environment with different folders and permissions.
Expanding AI also means cleaning up old permissions
Sentra was founded in 2021 by Yoav Regev, Asaf Kochan, Ron Reiter and Yair Cohen, and lists headquarters in New York and Tel Aviv. The investment adds resources where AI adoption meets an older organisational task: identifying where information lives, who owns it and which accounts genuinely need it.
For a security manager, a funding round is not a reason to change permissions immediately. It does indicate the product's direction: continuous management of information and identities beyond a one-off scan. Before connecting an agent to a business process, teams should define its necessary data sources, the permission owner and how to withdraw access if its behaviour changes.
That is where the broader promise of agent security is tested. Less unnecessary access leaves less room for error or misuse. Sentra's new investment is intended to expand that protection. Results for each customer will depend on repository coverage, classification accuracy and the permissions the organisation is prepared to fix.