AegisAI Feature
Classify data the same way every time
AegisAI, Data Classification
Example figures- CLASSIFIED
- deterministic
- same in, same out
- CLASSES
- PII · PHI · secrets
- + regulated
- DRIVES
- DLP + PDP
- sensitivity ladder
- Email + SSN in text
- PII
- API key pattern
- secret
- Patient record fields
- PHI
- Public marketing copy
- non-sensitive
- Engine: data_classification.py
- deterministic
Why classification is foundational
Almost every security decision depends on one question: how sensitive is this data? Whether a prompt should be gated, whether an action is a sensitive write, whether a tool-call crosses a data boundary: all of it needs a reliable sensitivity signal. AegisAI’s data classification engine (data_classification.py) provides it deterministically, so the label is consistent rather than a fresh guess each time.
What it labels
| Class | Examples | Where it drives a decision |
|---|---|---|
| PII | Names, emails, SSNs | DLP gating, sensitivity ladder |
| Secrets | API keys, tokens, credentials | Block on egress paths |
| PHI | Patient and health data | HIPAA-aligned handling |
| Financial | Account and card data | PCI-aligned handling |
| Non-sensitive | Public content | Allowed to flow |
How the signal is used
- The PDP’s sensitivity ladder uses it to distinguish read, write, and sensitive
- AI DLP uses it to decide which flows to gate
- Data-boundary policy uses it to decide which data may cross
- The Action Trust Score lowers trust when sensitive data is involved
One classifier, many controls
Because a single deterministic engine feeds DLP, the decision point, and the trust score, sensitivity means the same thing everywhere. You are not reconciling three different notions of “sensitive” across three tools: the classification is the shared vocabulary the whole platform reasons over.
Questions people ask
What is a data classification engine?
A data classification engine labels content by sensitivity: PII, secrets, PHI, financial, or non-sensitive. AegisAI’s engine does this deterministically, so the same content always gets the same label, which makes it a reliable input for DLP, the sensitivity ladder in the decision point, and the Action Trust Score.
Why does deterministic classification matter?
If a classifier can label the same content differently on different runs, any policy built on it is unenforceable. A deterministic engine gives a stable, repeatable label, so you can set a rule once and trust it applies consistently, and so an auditor can reproduce the classification.
How is classification used in decisions?
The classification feeds several controls: the decision point’s sensitivity ladder uses it to separate reads, writes, and sensitive actions; AI DLP uses it to decide what to gate; data-boundary policy uses it to decide what may cross; and the Action Trust Score lowers trust when sensitive data is involved.
What data classes does it recognize?
It recognizes common sensitive classes including PII, secrets and credentials, PHI, and financial data, alongside non-sensitive content. Those classes map to the handling expectations in frameworks like HIPAA and PCI DSS, so classification and compliance line up.
Keep reading
Guides
Calculators
One finance briefing a week
New guides, checklists and benchmarks, by email. Unsubscribe from any issue.
Give every control a sensitivity signal
Classify data deterministically so DLP and decisions rest on one consistent label.