Concentric AI adds vision model to spot sensitive files
Wed, 19th Aug 2026 (Today)
Concentric AI has introduced a vision model feature in its Semantic Intelligence platform to detect sensitive documents by their visual signatures.
The tool is designed to recognise items such as passport photo pages from any country and driver licences from any US state, even when text is blurred. Rather than relying on text extraction, it analyses a document's consistent visual characteristics to determine its type.
The approach addresses a longstanding problem in data security and governance. Companies often hold sensitive records across large, fragmented environments, with information stored in image files and other formats that can be difficult to classify using conventional methods.
Optical character recognition has long been used to process document images, but it depends on reading characters from pixels and converting them into machine-readable text. According to Concentric AI, that process can consume significant computing resources and may be less reliable when image quality is poor.
The company argues that visual structure can sometimes provide a stronger signal than text when identifying a document. This is especially relevant for standardised documents whose layout, formatting and design remain recognisable even when individual words or numbers are obscured.
Passports and driving licences are the immediate examples cited, but the underlying method also extends to organisation-specific records. Documents created for internal use can develop repeatable visual patterns, making them candidates for similar classification models.
OCR limits
The launch reflects a broader shift in security software towards analysing context and pattern as well as content. For companies trying to map where regulated or sensitive information sits, image-based files have remained a weak spot because they do not always yield accurate text for traditional scanning tools.
Low-quality scans, partial images and blurred content can all reduce the effectiveness of OCR-based systems. At the same time, security teams are under pressure to identify and protect these records because threat actors may still be able to extract useful information from imperfect images.
Concentric AI says the feature offers an alternative route to classification by focusing on document identity rather than trying to decipher every visible character. It is available immediately.
Partners will also be able to help customers build dedicated models for documents with consistent visual traits in their own environments. This points to an effort to extend the technology beyond common identity documents into internal records relevant to compliance or operational security.
Company view
Dr Madhu Shashanka outlined the rationale for the launch.
"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. As a leader in the field, Concentric AI is continuously advancing innovative approaches to data security challenges and leveraging signals which traditional approaches to discovery completely ignore, going beyond discovery of textual data to discover all sensitive information, no matter the format," Shashanka said.
The announcement places Concentric AI in a contested market, with vendors trying to improve how businesses find, classify and govern sensitive data across cloud services, internal systems and AI workflows. As more organisations adopt generative AI tools, understanding what data exists and where it resides has become more urgent.
Image files can be especially difficult to assess at scale because they often sit outside structured databases and may not include reliable metadata. Security teams therefore need methods that can identify risk without depending solely on file names, tags or extracted text.
The new feature is available now and can identify standardised sensitive documents through their visual consistency, even when text within those records cannot be read clearly.