Enterprises are required to protect all of the sensitive information they process and store, but the challenge in addressing data security risks is often maintaining visibility into the data that is sprawled across their environments in various formats.
Organizations need to understand what personal, financial, health and other regulated information they hold, where it is stored and who or what can access it. It’s a task that becomes more complicated as data spreads across cloud storage, collaboration platforms, applications, databases and unstructured files. Images and scanned documents can be particularly difficult to classify, as conventional systems often depend on extracting text before determining the contents of a document.
Concentric AI aims to address this by adding a vision model capability to its Semantic Intelligence platform that is designed to identify sensitive documents such as passport photo pages and driver licenses based on consistent visual characteristics. Documents that are specific to a particular company can also have identifiable characteristics.
The technology can identify a document type even when the contents of the image are blurred, enabling organizations to detect potentially sensitive files without relying solely on the text they contain. It uses fewer computing resources than optical character recognition (OCR), which processes each pixel in an image to detect and convert characters into machine readable text, requiring a high-quality input.
“Many sensitive documents carry a distinct visual identity,” said Dr. Madhu Shashanka, Co-Chief Technology Officer and Co-Founder at Concentric AI. “Not taking advantage of these visual cues leaves valuable signals untapped and limits the capability to discover and protect sensitive data.”
The new feature will allow Concentric AI’s partners to work with customers to develop dedicated models for identifying sensitive data in their environments, the company said.
AI Creates Another Layer of Data Risk
Enterprises face a growing set of data security challenges with the adoption of GenAI and AI agents, which can process large volumes of data and increase the risk of breaches enabled by poor classifications or excessive permissions that make sensitive information available to applications and models that do not need to access it.
Enterprises need to establish which data is sensitive, apply appropriate access controls and monitor how the data can be accessed or processed by AI systems as well as human employees.
Autonomous AI agents introduce further complexity because they can retrieve information and take action across multiple applications with limited human intervention.
That makes visibility and control over enterprise data increasingly important.
As Adam Spearing, EMEA Head of AI Innovation at ServiceNow, told CX Today in a separate discussion:
“When I go and access data, some of it I’m allowed, some of it I’m not. And you’ve got to have the same principles for your AI agents and for your AI.”
“How do we make sure it knows what data to go to? How does it know what systems it’s allowed to go into? How does it know what its function and role is?” Spearing added.
At the same time, the ability to use AI to interpret images and documents that can improve legitimate data discovery can also create security risks, as cyber attackers are increasingly using AI to automate social engineering and data extraction.
Concentric pointed to that issue, noting that while OCR can struggle to process information in low-resolution images accurately, cyberattackers could potentially use AI to successfully extract information for malicious purposes.
Data discovery will be key for security teams dealing with large collections of scanned or image-based documents, becoming part of a control layer that restricts an agent’s access to sensitive information.
The effectiveness of a tool like Concentric AI’s will depend on how accurately enterprises can classify the data they hold and connect that classification to broader identity management, data access, governance and monitoring controls.
“The context around the data is as important as the quality of the data when it comes specifically to AI,” Spearing said.
As AI systems gain access to larger volumes of enterprise data, visibility into the context of information needs to go beyond metadata and extracted text to determine what those systems should, and should not, be allowed to access.
