Concentric AI Wants to Help Enterprises Find and Secure Sensitive Data Hiding in Documents
18:43 · August 18, 2026 · CX Today
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 […]
Summary
Concentric AI has introduced a vision model within its Semantic Intelligence platform that classifies sensitive documents such as passport photo pages and driver licenses by recognizing consistent visual traits rather than extracting text. The approach works even when images are low-resolution or blurred and requires fewer computational resources than optical character recognition, which processes every pixel to convert characters into machine-readable form. Company-specific documents that share distinctive layouts can also be identified, and partners can train dedicated models for individual customer environments.
This capability addresses a long-standing gap in enterprise data discovery: unstructured and image-based files scattered across cloud storage, collaboration tools, and internal repositories often escape conventional classification systems. Without reliable identification of personal, financial, or regulated content, organizations cannot determine where sensitive material resides or who and what can reach it. The new model supplies an additional signal that complements existing metadata and text-based methods.
The need for improved visibility has become more urgent with the spread of generative AI and autonomous agents. These systems can ingest large volumes of data and act across multiple applications with limited oversight, amplifying the consequences of excessive permissions or incomplete classification. Enterprises must therefore map sensitive information, enforce role-appropriate access controls, and monitor both human and machine interactions. As Adam Spearing of ServiceNow has noted, the same authorization principles applied to employees must extend to AI agents so they reach only the data and systems required for their defined functions.
At the same time, the same AI techniques that aid legitimate discovery can be exploited by attackers for automated data extraction. Concentric AI’s emphasis on visual cues before text processing therefore forms part of a broader control layer that restricts agent access and supports governance frameworks. The ultimate effectiveness of such tools depends on how precisely classifications are linked to identity management, monitoring, and policy enforcement across the enterprise.
Why it matters
Directly addresses AI-driven data classification to mitigate security and privacy risks, actionable for Dutch teams handling GDPR and EU AI Act compliance. Provides practical guidance on controlling AI agent access to sensitive information, relevant for security and privacy professionals.
