AI READY Data Security
A technology that prepares data in a form AI can learn from and use, while keeping security attributes such as level, compartment, and distribution control, together with history records, attached to the data throughout the process.
Research and collaboration
We are seeking academic institutions to jointly develop AI READY data security with us.
Limits of existing approaches
Inconsistent data formats and descriptions
Field names, code values, and units differ by department and system, and descriptive documents are lacking, so existing data is hard to use directly for AI training.
General standards that assume open release
General public data standards assume open release, so the release and use of defense and defense industry data mixed with non-public or sensitive company information can be restricted altogether.
Security attributes separated during preparation
During refinement and conversion, the original's level markings and access rights are not carried over to new files, and it becomes hard to confirm who later used data of which level.
Existing approaches and AI READY data security
This summarizes general tendencies and may differ by product and environment.
Data standardization
Field names, code values, units, date formats, identifiers, schemas, and notation for missing and special values are converted according to the organization's standardization rules and exported in formats used for AI training, such as CSV, JSON, and Parquet.
The correspondence between the original and the converted result is recorded in the refinement and conversion history.
Quality assessment and de-identification
Quality indicators such as missing values, duplicates, format errors, and value ranges are assessed, and the results are kept with the data.
Personal information and sensitive company information are included in training data after pseudonymization or anonymization, and the processing criteria and affected fields are recorded with them.
AI READY data package composition
The data itself, metadata, a data card, quality assessment results, refinement and conversion history, a usage guide, and security and license information are composed into a single package.
The data card and usage guide summarize the data's source, composition, scope of use, and restrictions, so the receiving party can check the conditions of use without separate documents.
Each component of the package is managed by version.
Security attributes that move with the data
Security attributes such as level, compartment, and distribution control are contained in data files and packages, so they are not separated from the content even when files are copied or moved to AI training and usage environments.
Refined and converted outputs saved in the secure area are automatically assigned at least the default level set by the organization.
See data-centric securityConditions of use and control of AI usage paths
Only users who meet all three conditions of level, compartment, and distribution control as well as the license conditions can open the data. Data provided externally is delivered as a distribution copy with designated recipients and a validity period, and can be recalled after it is provided.
AI agents cannot access protected data directly. Their requests go through a mediation step, and only content within the lower of the user's clearance and the organization's limit is passed on.
See AI agent securityRefinement and conversion history and usage history
The processing performed and the person who performed it are recorded at each step of standardization, quality assessment, and de-identification, so it is possible to confirm from which original and through which processes training data was created.
Views, prints, exports, level changes, blocks, and AI agent requests are recorded in the data access history with the user, time, and target, and data security posture management shows the level distribution, unprotected data, and exposure of protected data.
Assurances
Alignment with AI READY data management aspects
Addresses the eight management aspects of AI READY data.
Data preparation and security aspects are managed together within a single data package.
Try JOLDISK for free
