OPSWAT AI Content Inspector release notes
V 2.3.1
Release date: 29 September 2026
Release type: MINOR
AI Content Inspector 2.3.1 adds support for WMF/EMF images and OpenDocument spreadsheets and presentations, introduces a Linux NVIDIA GPU engine package, and delivers updated image and text detection models, revised default confidence thresholds, and stability fixes for e-mail, Office, and text processing.
New Features
Added support for WMF and EMF images (.wmf, .emf)
Added support for OpenDocument spreadsheets and presentations (.ods, .ots, .odp, .otp)
Added a Linux NVIDIA GPU engine package (requires CUDA 13.0 and NVIDIA driver R580 or newer)
Added engine and database versions to every scan result
Added metadata forensic analysis for images attached to .eml and .msg messages
Changes and Improvements
Updated the ML SDK from 1.7.0 to 1.8.0, model weights from 1.7.0 to 1.8.0, and the DB API from 10 to 11
Replaced OriginLens v0.5.1 with v1.4.2 in both FAST and FORENSIC profiles. The CNN and CommFor detectors and image heat maps have been removed;
output_imagesremains available in the Core interfaceReplaced the RoBERTa text classifier with Ettin-17m in the FORENSIC profile, and added Ettin-17m as a second text verdict source in the FAST profile
Updated default confidence bands: images now use 60/40 (previously 70/30), and the text suspicious threshold is now 50 (previously 45). Workflows still using the previous defaults are migrated automatically
Texts shorter than 45 words are no longer scored and are reported as Uncertain
Python engine modules are now compiled with Nuitka
Bugfixes
Fixed an issue where scanning .msg files with certain code pages failed with error code 1
Fixed an issue where correctly typed .msg files were rejected as unsupported
Fixed an issue where capture date, camera, and GPS values stored in nested EXIF IFDs were not read
Fixed an issue where the fraud result was reported as
not detectedwhen the fraud classifier did not run; it is now reported asnot scannedFixed an issue where unreadable Office documents were reported as an unsupported file type; they now fail with
err_code1 and a descriptive error messageFixed an issue where files with no extractable text returned a model score; they now report text as Not Detected
Fixed an issue where text scans did not fail when the required text classifier could not be loaded
V 2.3.0
Release date: 14 September 2026
Release type: MAJOR
AI Content Inspector 2.3.0 introduces support for Microsoft Office document formats, adds engine lifecycle notifications, and includes improvements to document text processing, an ML model update, and fixes for result consistency.
New Features
Added support for Microsoft Word documents (.docx, .docm, .doc, .rtf, .odt)
Added support for Microsoft PowerPoint presentations (.pptx, .pptm, .potx, .potm, .ppsx, .ppsm, .ppt, .pot, .pps)
Added support for Microsoft Excel spreadsheets (.xlsx, .xlsm, .xltx, .xltm, .xlam, .xlsb, .xls, .xlt)
Added out-of-support lifecycle notifications, informing users when the engine version reaches end of support
Changes and Improvements
Reworked document text cleaning to improve the quality and consistency of text extracted for analysis
Improved the performance of markdown text cleanup
Bugfixes
Fixed an issue where the result JSON could be inconsistent when preferences were misconfigured
Fixed an issue where sub-verdicts could be inconsistent when scanning with some features disabled
V 2.2.0
Release date: 14 August 2026
Release type: MAJOR
AI Content Inspector 2.2.0 introduces support for additional email file formats, expands forensic analysis capabilities for embedded images, and includes improvements to provenance validation, image processing, and overall engine stability.
New Features
Added support for E-mail Message (.eml) files
Added support for Microsoft Outlook Message (.msg) files
Added metadata forensic analysis for images embedded within PDF documents, enabling the engine to inspect image-level metadata and forensic signals extracted from PDF content
Changes and Improvements
Strengthened C2PA provenance validation and handling to improve reliability when analyzing Content Credentials and provenance information
Improved image input normalization to provide more consistent preprocessing across supported image formats and improve downstream analysis reliability
Bug Fixes
Fixed a dependency issue affecting the C2PA library
Fixed processing issues affecting EPS files
Fixed an issue where grayscale and bilevel EPS images could fail during processing
Resolved a stability issue where processing FPX files could cause the scan to terminate unexpectedly. FPX is no longer supported by AI Content Inspector
Fixed an issue affecting LA-mode images, including certain
.curcursor files, which could fail with a"not enough image data"error
V 2.1.1
Release date: 10 August 2026
Release type: MINOR
Detect AI Model Family from Image Metadata
New OriginLens image detector
Improved speed for image scanning (fast & forensic mode)
Google SynthID watermark detection for images is now reported under its proper name, consistent with the text watermark detector
Text detector display names aligned with the detection SDK
Fixed legitimate high-resolution images failing with "Image size exceeds limit … could be decompression bomb". The limit is now 250 megapixels by default and configurable via AIGCD_MAX_IMAGE_PIXELS
Fixed image fingerprint detection failing on files with malformed XMP metadata
Fixed scans aborting when an embedded image inside a document could not be decoded
Fixed Stable Diffusion 3 provenance being reported under an inconsistent model name
V 2.1.0
Release date: 13 July 2026
Release type: MAJOR
Introduce Fast mode and Forensic mode
Fast mode: Optimized for high-speed analysis with a fast response
Forensic mode: Optimized for deep analysis and investigative workflows
Added an option to disable output image generation
Improved PDF extraction performance
Improved scan performance with faster processing times
V 2.0.1
Release date: 28 May 2026
Release type: MINOR
Upgraded the core image and text detection components for improved accuracy and performance
Resolved issues affecting text scanning functionality
Enhanced support for offline and packaged deployments
V 2.0.0
Release date: 21 May 2026
Release type: MAJOR
Added 10 new image detectors to analyze AI-generated footprints in visual content
Introduced 9 new text detectors to identify AI-generated footprint signals in textual content
Enhanced results page with deeper insights, providing more detailed outputs from individual detectors, including images and analysis breakdowns
Introduced a new Uncertain Hits Threshold setting to improve control over uncertain detection results
Added support for PDF file type including both AI text detection and AI image detection workflows
Introduced fraud detection support for the following content types:
Invoice fraud
Car accident insurance fraud
Home insurance fraud
Introduction
OPSWAT AI Content Inspector 2.0.0 marks the General Availability release of OPSWAT’s proprietary AI-powered content authenticity and document fraud detection engine. Purpose-built for modern, high-throughput environments, AI Content Inspector helps organizations identify AI-generated content and detect AI fraudulent documents with speed and accuracy.
Fully Integrated with the MetaDefender Platform
AI Content Inspector is fully integrated into the MetaDefender platform and is available across MetaDefender products and deployment environments, including both cloud and on-premises installations on Windows and Linux.
Working alongside Proactive DLP and Deep CDR, AI Content Inspector extends content authenticity and fraud detection coverage through advanced AI technologies. The engine analyzes both textual and visual artifacts to deliver AI-generation and fraud-indicator verdicts, enabling organizations to make informed security decisions before content enters critical workflows.
Designed to Combat Modern Fraud
AI Content Inspector focuses on the primary entry points for content-based fraud across enterprise workflows. It supports a wide range of image formats, text-bearing documents, and PDF files where AI-driven deception and document fraud most commonly occur, including:
Invoice and accounts payable fraud
AI-generated business and financial documents
AI-generated or manipulated home damage images and insurance claims
Detection of AI-generated insurance claim evidence
Key Benefits
Accelerated Content Review
Faster review cycles for high-volume workflows such as claims processing and invoice verification
Improved operational efficiency while reducing false positives
Intelligent Pre-Decision Analysis
Rapid content inspection enables organizations to automatically allow, flag, block, or route content for further review at scale
Comprehensive Fraud Detection Coverage
Detects both AI-generated content and document fraud indicators across common enterprise fraud scenarios, including:
AI-generated invoices and financial documents
AI-generated vehicle and accident imagery fraud
AI-generated home and property insurance fraud
AI-generated synthetic claim documentation
Broad File Format Support
Extensive support for image formats commonly used in fraud attempts
Coverage for text-bearing documents and PDF files across enterprise workflows
Availability
AI Content Inspector 2.0.0 is available as a standalone engine within MetaDefender Core and MetaDefender Cloud, providing organizations with advanced AI-driven content authenticity verification and document fraud detection capabilities across their security ecosystem.