VERIFIED AI
AI Transparency Disclosure
verifiedai.app/transparency · Last Updated: March 14, 2026
About This Disclosure
This page provides transparency about the artificial intelligence systems and training data used by Verified AI, a product of Terminal Labs LLC. This disclosure is made in compliance with the California Generative Artificial Intelligence Training Data Transparency Act (AB 2013), effective January 1, 2026, and reflects our commitment to transparency in AI development.
This disclosure is updated at least annually, or whenever a substantial modification is made to our AI systems. The "Last Updated" date above reflects the most recent revision.
1. System Overview
Purpose and Intended Use
Verified AI is an AI-powered luxury goods authentication platform. Our system analyzes user-submitted photographs of consumer products (such as handbags, sneakers, watches, and other luxury items) and generates a statistical probability of authenticity. The system is designed to serve as a supplemental informational tool and does not provide professional appraisals, legal certifications, or guarantees of authenticity.
How the System Works
Users upload photographs of items through the Verified AI mobile application or website. Our AI models analyze visual characteristics of the items, including materials, construction details, logos, stitching patterns, hardware, and other identifiable features. The system compares these characteristics against learned patterns from its training data to produce an authenticity assessment score.
Known Limitations
Our AI system has inherent limitations, including but not limited to:
- The system analyzes digital images only and cannot detect tactile, olfactory, weight-based, or internal construction characteristics.
- Results may be affected by image quality, lighting conditions, angle, resolution, and digital compression.
- High-quality counterfeits ("super-fakes") may not be reliably detected through photographic analysis alone.
- The system may produce both false positives (identifying a counterfeit as authentic) and false negatives (identifying an authentic item as counterfeit).
- The system's accuracy varies by product category, brand, model, and the era in which an item was manufactured.
For high-value items or transactions with significant financial implications, we recommend consulting a qualified human authentication expert in addition to using our Service.
2. High-Level Summary of Training Datasets
The following is a high-level summary of the datasets used in the development and training of Verified AI's AI models, as required by California AB 2013.
2.1 Dataset Sources and Descriptions
Authenticated Product Image Datasets
- Source: Proprietary datasets compiled by Terminal Labs LLC from licensed and authorized sources, including authenticated product photographs obtained from authorized resellers, consignment platforms, and authentication professionals.
- Description: Photographs of verified authentic luxury goods across multiple product categories (handbags, footwear, watches, accessories). Images include various angles, lighting conditions, and levels of detail.
- Purpose: These datasets train the model to recognize visual patterns, construction details, and design characteristics consistent with authentic products.
Known Counterfeit Product Image Datasets
- Source: Proprietary datasets compiled by Terminal Labs LLC from licensed and authorized sources, including known counterfeit items identified by authentication professionals, customs seizure records (where publicly available), and partnerships with authentication service providers.
- Description: Photographs of verified counterfeit products across the same product categories, enabling the model to learn distinguishing features between authentic and counterfeit items.
- Purpose: These datasets train the model to identify visual inconsistencies, manufacturing defects, and patterns commonly associated with counterfeit goods.
Publicly Available Product Reference Data
- Source: Publicly available product images, catalog data, and reference materials from brand websites, retailer websites, and publicly accessible databases.
- Description: Reference imagery used to supplement training data with additional examples of authentic product designs, colorways, seasonal variations, and model-specific details.
- Purpose: To broaden the model's coverage across product lines, release dates, and regional variations.
2.2 Data Collection Methods
- Direct capture: photographs taken under controlled conditions of authenticated items.
- Licensed acquisition: datasets obtained through licensing agreements with authentication service providers and resale platforms.
- Public collection: images sourced from publicly available websites and catalogs where permissible under applicable terms of use.
Data collection is ongoing to incorporate new product releases, seasonal collections, and emerging counterfeit techniques.
2.3 Number of Data Points
Our training datasets include imagery across multiple product categories. The approximate scale of our datasets is as follows:
- Total images: Tens of thousands of product images across all categories (exact figures are proprietary).
- Product categories covered: Handbags, sneakers/footwear, watches, and accessories.
- Dataset type: Dynamic. New data is incorporated on an ongoing basis as new products are released and as our authentication coverage expands.
2.4 Intellectual Property Status of Training Data
- Copyrighted material: Some training images may depict products that are subject to third-party trademark or design protections. Our use of such images for the purpose of training AI models for authentication analysis is conducted in accordance with applicable law, including the fair use doctrine under U.S. copyright law.
- Public domain data: Certain reference images and product catalog data used in training are in the public domain or are made available under open or permissive licenses.
- Licensed data: Portions of our training data are obtained under licensing agreements with data providers, authentication professionals, and resale platforms. The terms of these agreements are confidential.
2.5 Personal Information
Our training datasets do not contain personal information as defined by the California Consumer Privacy Act (CCPA). Training data consists of photographs of items and products, not of identifiable individuals. We do not use biometric data, facial imagery, or other personally identifiable information in our training datasets. If user-submitted Content is used for model improvement, it is first aggregated and de-identified so that it cannot reasonably be used to identify any individual user.
2.6 Synthetic or AI-Generated Training Data
We may use limited quantities of synthetic (AI-generated) data to augment our training datasets. Synthetic data is used to simulate lighting conditions, angles, and image quality variations to improve model robustness. Synthetic data is clearly labeled in our internal dataset management systems and is not used to create fabricated examples of counterfeit goods.
2.7 Data Cleaning and Processing
Training data undergoes quality review and processing before use, including removal of duplicate images, filtering of low-quality or unusable photographs, normalization of image formats and resolutions, and manual or semi-automated verification of authenticity labels by qualified reviewers.
3. Model Information
Model Architecture
Verified AI uses proprietary computer vision models based on deep learning architectures, including convolutional neural networks (CNNs) and/or vision transformer architectures. Specific architectural details are proprietary and constitute trade secrets of Terminal Labs LLC.
Training and Evaluation
Models are trained, validated, and tested using standard machine-learning practices, including train/validation/test data splits, cross-validation, and performance evaluation against held-out test sets. We continuously evaluate model performance across product categories to identify and address accuracy gaps.
Third-Party AI Providers
Certain components of the Service may utilize AI infrastructure or models provided by third-party providers. Where third-party AI models are used, they are subject to contractual data processing and confidentiality obligations. User Content submitted for analysis is processed in accordance with our Privacy Policy.
4. Fairness and Bias Mitigation
Verified AI's models analyze characteristics of physical products, not characteristics of individuals. Our system does not make decisions based on, and is not designed to assess, any characteristic of any person, including race, gender, ethnicity, national origin, age, disability, or any other protected class.
We monitor model performance across product categories and price points to identify potential disparities in accuracy. If we identify significant performance gaps between categories, we take steps to supplement training data and improve model coverage.
If you believe an assessment reflects bias or inaccuracy that may be attributable to a systematic issue, please contact us at privacy@verifiedai.app.
5. User Data and AI Model Training
We do not use your personal information or identifiable Content to train our AI models without your explicit, affirmative consent.
We may use aggregated and de-identified data derived from user interactions to improve our models. Aggregated and de-identified data cannot reasonably be used to identify any individual user.
If you wish to opt in to contributing your Content for model training, you may do so through your Account Settings. You may withdraw consent at any time, and any withdrawal will apply prospectively. For more information, see our Privacy Policy.
6. Human Oversight
Verified AI is an automated AI tool. Assessments are generated by machine-learning algorithms without direct human review of individual results, unless otherwise expressly stated. We recommend that users treat assessments as one input among several when making purchasing or selling decisions, and that users consult a qualified human expert for high-value or high-stakes transactions.
7. Contact Information
For questions about this disclosure, our AI systems, or our training data practices, please contact:
Terminal Labs LLC
Email: privacy@verifiedai.app
Website: verifiedai.app
This disclosure is provided for informational and compliance purposes. It does not alter or supersede the Verified AI Terms of Service or Privacy Policy.