01
Public
“Proofread our launch announcement.”
Allow
02
Internal
“Summarize today’s team meeting.”
Allow and log
03
Confidential
“Draft a reply using this contract clause.”
Warn and coach
04
Restricted
“Why does this call fail with our live key?”
Block
Overview
Wald AI provides context-aware data classification for AI assistants.
Traditional Data Loss Prevention (DLP) systems rely on pattern matching techniques such as regular expressions to identify sensitive information. While effective for structured identifiers like credit card numbers or Social Security numbers, these systems:
- Generate high false positives
- Miss business-critical confidential data expressed in natural language
Wald AI takes a different approach.
Wald AI’s classification engine uses semantic understanding models to identify sensitive data based on meaning and context, not just patterns.
This enables:
- Detection of sensitive information beyond structured formats
- Significant reduction in false positives
Key capabilities include:
- Context-aware sensitive data detection
- Multi-label classification of data types
- Fine-grained taxonomy of sensitive data categories
- Integration with enterprise DLP and governance framework.
| Traditional Data Classification Types |
Wald AI’s Context-Aware Classification |
| Pattern matching (regex) |
Semantic understanding |
| High false positives |
Dramatically reduced false positives |
| Misses natural language data |
Detects confidential information in context |
| Single-label detection |
Multi-label classification |
| Requires constant rule updates |
Adapts to meaning and context |
What Is Data Classification?
Data classification is the process of labeling information by how sensitive it is, so your team knows who can access it, where it can go and how it must be protected. Small teams only need a short list of levels everyone can apply.
What Are the Four Data Classification Levels?
Most teams use four levels, and the table shows what each one looks like in an AI prompt.
| Level |
Examples |
Example AI prompt |
Policy action |
| Public |
Published blog posts, press releases |
“Proofread our launch announcement.” |
Allow |
| Internal |
Meeting notes,
process documents
|
“Summarize today’s team meeting.” |
Allow and log |
| Confidential |
Contracts,
salary data,
customer lists
|
“Draft a reply using this contract clause.” |
Warn and coach |
| Restricted |
PII,
API keys
|
“Why does this call fail with our live key?” |
Block |
How to Classify Data in Five Steps
- Start with sensitive data discovery: list where sensitive data lives across drives, chat tools and code repositories.
- Adopt the four levels above and write a one-line definition for each.
- Label every data type with a level and an owner.
- Set one handling rule per level, including what may go into AI tools.
- Review quarterly, because new tools and prompts create new data exposure.
Identity Data
Information that directly identifies an individual.
Name Data
Names or aliases that identify a specific individual.
Examples:
- Jane Doe
- Dr. Michael Patel
- Anya Sharma
Context:
Please send the contract to Dr. Michael Patel.
National ID Data
Government-issued identification numbers tied to individuals.
Examples:
- US Social Security Number
- UK National Insurance Number
Context:
My Social Security number is required for verification.
Passport Data
Identifiers associated with passports.
Examples:
- Passport number
- Issuing country
- Expiration date
Context:
Passport number 123456789 expires in 2029.
Driver’s License Data
Government-issued driver license identifiers.
Examples:
- License number
- License class
- Issue date
- Expiration date
Context:
License number F1234567 needs renewal.
Phone Number Data
Telephone numbers including mobile and landline.
Examples:
Financial Data
Financial identifiers and records associated with accounts or transactions.
Credit Card Data
Payment card information.
Examples:
- 4111 1111 1111 1111
- CVV
- Expiry date
Context:
My credit card number is 4111 1111 1111 1111.
Bank Data
Bank account identifiers.
Examples:
- Account numbers
- Routing numbers
- SWIFT codes
Context:
Account number 123456789
IBAN Data
International bank account identifiers.
Examples:
- GB29 NWBK 6016 1331 9268 19
- DE89 3704 0044 0532 0130 00
Financial Records
Sensitive financial information.
Examples:
- Salaries
- Portfolio values
- Financial reports
Context:
The executive compensation package is $450,000 annually.
Healthcare Data
Health-related information tied to individuals.
Medical Data
Medical conditions, treatments, or clinical terms.
Examples:
- Type 2 Diabetes
- Appendectomy
- Lisinopril
Health Identifiers
Identifiers used in healthcare systems.
Examples:
- Medical Record Number (MRN)
- Patient ID
Context:
Patient MRN 123456789
Insurance Data
Insurance-related identifiers.
Examples:
- Policy number
- Group ID
- Claims history
Corporate Data
Information related to organizations and internal operations.
Company Data
Identifies organizations or internal initiatives.
Examples:
- Verizon Communications
- The Mayo Clinic
- Project Atlas
Context:
Draft an email announcing Project Phoenix.
Product Data
Details about products or services.
Examples:
- Product roadmaps
- Feature specifications
- Release timelines
Context:
Next-generation EV model roadmap.
Intellectual Property Data
Confidential or proprietary knowledge.
Examples:
- Unpublished research
- Internal datasets
- Research notes
Employment Data
Information related to roles, performance, or workforce activity.
Examples:
- Performance reviews
- Resume content
- Internal role changes
Technical and Security Data
Identifiers used in systems, infrastructure, and security.
API Credentials
Identifiers used for integrations.
Examples:
- AWS Access Key ID
- OAuth client ID
- Google Cloud Client ID
Secret Keys
Private credentials for authentication or encryption.
Examples:
- AWS Secret Access Key
- SSH private key
- JWT signing secret
Hardware Identifiers
Device-level identifiers.
Examples:
- MAC address
- IMEI
- Serial number
Internet Identifiers
Online communication identifiers.
Examples:
- IP addresses
- Email addresses
- Domain names
- Usernames
Personal Attributes
Characteristics describing an individual.
Demographic and Identity Attributes
Examples:
- Gender
- Ethnicity
- Nationality
- Political affiliation
- Sexual orientation
Context:
She was born in Mexico.
Behavioral and Activity Data
Information generated from user actions and interactions.
Behavioral Data
Examples:
- Browsing history
- Clickstream data
- Search history
Geolocation Data
Precise location information.
Examples:
- 40.7128° N, 74.0060° W
- GPS location history
Contextual Detection Differentiators
Wald AI’s core advantage lies in contextual interpretation.
Date Disambiguation
- My card will expire on 01/25 → Credit Card Data
- My license is expiring on 01/28 → Driver’s License Data
Ambiguous Identifiers
Same number, different meanings:
- Account number 123456789 → Bank Data
- SSN 123-45-6789 → National ID Data
- Passport No. 123456789 → Passport Data
- Patient MRN 123456789 → Healthcare Data
- Order number 123456789 → Non-sensitive transactional data
Credit Card vs Order Number
- 378282246310005 (with payment context) → Credit Card Data
- Order number 378282246310005 → Non-sensitive data
Wald AI Data Classification Taxonomy
Wald AI uses a hierarchical taxonomy that fits alongside common security frameworks.
It extends traditional classification systems to support:
- Company data
- Product data
- Financial transactions
- HR interactions
- Other unstructured enterprise data
This reduces false negatives that pattern-based systems typically miss.
Top-level categories:
- Identity Data
- Financial Data
- Healthcare Data
- Corporate Data
- Technical and Security Data
- Personal Attributes
- Behavioral and Activity Data
Why This Matters for CISOs
Reduced False Positives
- Fewer blocked workflows
- Better user adoption
- Lower operational overhead
Higher Detection Accuracy
Wald AI detects sensitive data in:
- Natural language
- Unstructured prompts
- Mixed-content inputs
Better Policy Control
- Fine-grained classification
- Role-based enforcement
- Audit-ready governance
Contextual Disambiguation
The same value can represent different data types depending on context.
Example: 123456789
- Checking account number → Bank Data
- SSN → National ID Data
- Patient MRN → Healthcare Data
- Order number → Non-sensitive data
Reduction of False Positives
Pattern-based systems often misclassify.
Example:
378282246310005 → flagged as credit card by regex
But in context:
Your order number is 378282246310005 → correctly classified as non-sensitive
This significantly improves real-world usability.
Integration with Enterprise Data Governance
Wald AI integrates with existing classification frameworks such as:
- Microsoft Purview
- PAN-based taxonomies
- Internal enterprise classification standards
Organizations can define custom policies using Wald AI’s classification outputs.
Our View: Classify the Prompt, Not Just the File
A label belongs to a file, but a prompt is new content: a document marked internal can carry restricted details once a paragraph is pasted into an AI tool. Manual policies can only address known behaviors, but AI interactions are dynamic and constantly evolving, so by the time a new exposure risk is discovered, it may be too late to configure a policy for it. That is why we classify the meaning and intent of each prompt with contextual classification, using multiple small language models (SLMs) on the device.
Summary
At Wald AI, we take a context-driven approach to sensitive data classification, built specifically for AI workflows. Because our classification reads the context around the data instead of relying on rigid pattern matching, it delivers higher detection accuracy with far fewer false positives. It also produces richer classification signals, which give your team stronger AI governance controls.