Concentric Inc. launched today with $7.5 million in a Series A funding and a new approach to document-level security.

The approach relies upon the artificial intelligence technique of deep learning to identify documents that are unprotected or inappropriately shared by analyzing their contents. The company was founded by an executive team with extensive experience at networking and security firms that include Juniper Networks Inc., PGP Corp., Symantec Corp., Hewlett Packard Enterprise Inc., Aruba Networks Inc. and Andiamo Systems Inc.

It uses a technology called Semantic Intelligence to scour an organization’s servers and classify documents according to their contents. The software then identifies access rights for those documents and automatically adjusts privileges based upon policies and permissions for similar documents. Actions can range from disabling an individual user’s access to changing group permissions.

Concentric estimates that the average business has 10 million documents ,of which about 1.2 million are critical to the business. It figures a 20% of those files carry improper sharing privileges, most of which are overly generous.

Its research is backed up by analysis published by competitor Varonis Systems Inc., which reported in 2018 that the average company it examined left 21% of its folders accessible to every employee. The problem is often caused by human error and can be almost impossible to spot without rigorous manual analysis.

The approach Concentric has developed uses contextual analysis to infer the contents of documents and identify those that are mostly likely to be sensitive.

“We build up semantic groupings of, say, contracts and design documents. We’re then able to label these clusters to assign category definitions to a particular customer,” said Chief Executive Karthick Krishnan. “You can determine whether or not the cluster grouping is sensitive and then conduct risk analysis to look, for example, for documents that have been marked public when others in the cluster are confidential.”

The company says it software requires no training because it uses language models that have been applied in other scenarios for years. A certain amount of human intervention is typically required in the early stages to verify classification choices and define exceptions, but Concentric’s platform learns over time to identify common characteristics and work with less supervision.

Although exceptions exist, most documents can be classified automatically, Krishnan said. “It’s very unlikely that if two documents are highly similar they will have different properties,” he said.

Krishnan said the technology can scan 6 million to 10 million files within a couple of weeks. The product works with popular document formats and document management systems including Microsoft Corp.’s Windows, Office365, OneDrive and SharePoint Online; Google LLC’s Drive; Box Inc. and Dropbox Inc. cloud file shares; and Amazon Web Services Inc.’s S3 object storage.

Funding was provided by Clear Ventures Inc., Engineering Capital LLC, Homebrew Management LLC and Core Ventures Group LLC. Krishnan said 70% of the funding will go into product development.

Photo: Morguefile