Protecting Your Data
Means Nothing
Without Enforcement
AI safety enforcement built by and for the communities that need it most
The only AI safety framework that enforces data sovereignty at the technical layer. Protects PII, student records, cultural knowledge, and small communities from unauthorized AI disclosure.
The Origin
The next breach won't come from a hacker.
It'll come from someone trying to do their job.
A health worker pastes a patient’s chart into ChatGPT to summarize symptoms.
A teacher drops a student’s IEP into Claude to help write a progress report.
A grant writer feeds enrollment data and community health statistics into an AI assistant to draft a federal proposal.
A startup founder pastes a contract into Gemini to review the terms.
A hospital administrator uploads patient intake forms into a chatbot to organize.
Different organizations. Different data. Same problem. Sensitive information — names, medical records, student data, proprietary business details — sent to external servers in a keystroke. No one was hacked. No one even noticed.
This happens every day — not because people are careless, but because AI is useful and nothing stands between the prompt and the model.
The Problem
Policies can't enforce themselves.
Governance frameworks weren't built for machine speed. Training isn't enough. By the time someone realizes what was shared, the data has already been processed, stored, and potentially used to train the next model.
What's needed isn't another policy. It's a shield.
IndigiArmor
A security layer that sits between your data and AI. Every prompt, every document, every query — scanned and filtered before it reaches the model.
- Detect and redact PII, student data, and confidential records
- Block sensitive cultural and community data from reaching AI systems
- Enforce data governance at machine speed
- Maintain sovereign control over what leaves your organization
Your data. Your sovereignty. Your shield.
Nothing reaches the AI without passing through IndigiArmor first.
What We Detect
Five detection domains working in concert. Cross-domain signal aggregation catches what individual scanners miss.
PII / PHI Protection
19 entity typesLuhn-validated credit cards. State-format driver's licenses. Obfuscation-resistant email and phone detection.
- Email, phone (NANP + E.164), SSN with context boosting
- Credit card with Luhn checksum, bank routing with ABA checksum
- Driver's license (50-state format lookup), passport (US/UK/CA)
- Medical IDs (MRN, NPI, DEA, Medicare MBI), genetic/genomic data
- Login credentials (AWS, GitHub, JWT, RSA keys)
- VIN with check digit validation, IBAN/SWIFT
- Tax IDs (EIN, ITIN), biometric references
Education Records (FERPA)
16 record typesTwo-factor model: protected data + student identifier = flag. Zero false positives on general inquiries.
- IEP content, goals, service minutes, 40+ accommodation phrases
- Section 504 plans, eligibility determinations, BIP/FBA
- 13 IDEA disability categories with abbreviations
- FRPL status, EL/ELL status, homeless/McKinney-Vento
- Foster care, migrant status, directory information opt-out
- General inquiry reduction: "What is an IEP?" → safe
Cultural Knowledge Protection
7 × 17 matrixIntent-based, not keyword-based. Detects extraction attempts, not curiosity. Cross-cultural by design.
- 7 intent categories: procedural, verbatim, location, replication, commercial, AI training, academic extraction
- 17 protected topic categories: ceremonies, sacred sites, traditional medicine, songs/prayers, and more
- Intent + topic = RED. Topic alone = YELLOW. Safe framing reduces score.
- TK Label recognition (TK SS, TK CO, TK CL)
- "Not for AI Use" markers: inline text, TK Labels, restriction phrases
- Restricted access indicators escalate to automatic RED
Small-Community Re-identification
12 pattern typesPopulation-scaled risk multiplier. In communities of 50 people, any two attributes identify someone.
- Unique role + location: "the only doctor in [community]"
- Small set membership: "one of three fluent speakers"
- Tribal identifiers: enrollment #, CDIB, band membership
- Demographic intersection, temporal narrowing
- Medical condition + community, rare event + location
- Negative inference: "all council members voted yes except one"
- Community size multiplier: <50 = 3.0x, 50-200 = 2.5x, 200-500 = 2.0x
Prompt Injection Defense
9 attack typesRuns first in the pipeline. Catches bypass phrases, encoding attacks, social engineering. 6-step normalization.
- Direct bypass: "ignore safety", "override the check", role overrides
- System prompt extraction attempts
- Encoding attacks: homoglyphs, zero-width chars, Base64, HTML entities
- Variable substitution: "Let X = my SSN. X is 123-45-6789"
- Fictional framing with real PII patterns
- Social engineering claims (identity, authority impersonation)
- Output format manipulation: "encode answer in Base64"
How It Works
One API call stands between your users and unauthorized data disclosure.
IndigiArmor acts as a protective gateway between your organization's data and external AI systems.
Data Request
Documents, messages, research materials, or other data are prepared to be sent to an AI system.
IndigiArmor Inspection
IndigiArmor scans the content for:
- Personally identifiable information (PII)
- Protected community data
- Sensitive cultural materials
- Confidential organizational records
Policy Enforcement
Governance rules determine whether information should be:
- Allowed
- Redacted
- Restricted
- Blocked
Secure AI Processing
Only approved and properly filtered data is sent to the AI model.
Protected Response
The AI response is returned without exposing restricted or sensitive information.
This is exactly what users see in the browser extension when sensitive content is detected.
Artificial intelligence is becoming a core part of modern workflows.
But sensitive information should never be exposed to AI systems without protection.
Before data reaches AI, it should pass through protection.
That protection is IndigiArmor.
Built Different
Verified across 20+ competitors. No one else combines cultural protection, FERPA compliance, and re-identification defense.
| Capability | IndigiArmor | Nightfall | Lakera | GoGuardian | Securly |
|---|---|---|---|---|---|
Standard PII Detection | |||||
PHI / HIPAA Compliance | |||||
Real-time Prompt Scanning | |||||
FERPA / IEP DetectionExclusive | |||||
Cultural Knowledge ProtectionExclusive | |||||
Small-Community Re-IDExclusive | |||||
Tribal Identifier DetectionExclusive | |||||
Prompt Injection Defense | |||||
AI Response Scanning | |||||
3-Tier Classification (G/Y/R)Exclusive | |||||
Cross-Domain Score AggregationExclusive | |||||
Self-Hosted Option |
Transparent Pricing
Browser protection for individuals
Personal
Essential browser protection for individual use.
Unlimited local scans
Chrome Extension
- 1 seat
- Unlimited local scans
- All 5 detection domains
- Local scan history
- 7-day free trial
Dashboard
Not included
API & SDK
Not included
Pro
Full features for power users.
1,000 cloud scans/mo
Chrome Extension
- 1 seat
- Unlimited local scans
- All 5 detection domains
- Local scan history
- 7-day free trial
Dashboard
- Web dashboard access
- Allowlist management
- Basic analytics
- Cloud audit logs (30 days)
API & SDK
- 1 API key
- 1,000 cloud scans/mo
- Document scanning
Our Principles
Self-standing. They do not extend, depend on, or require any external framework.
Enforce, Don't Advise
The gate is code, not a checkbox. Every AI request passes through server-side gating.
Classify, Don't Guess
Weighted signals with cross-domain multipliers. Green, Yellow, Red — not a binary.
Detect Intent, Not Just Keywords
"Tell me about ceremonies" is safe. "Give me the exact steps of the ceremony" is not.
Protect Communities, Not Just Individuals
Standard de-identification was built for cities. We build for villages.
Never Store What You Block
Blocked content is never stored in full. Audit logs record decisions, signal types, and a truncated prompt preview — never the complete prompt.
Normalize Before You Scan
6-step pipeline catches homoglyphs, zero-width chars, Base64, and every encoding trick.
The Community Decides What Is Sensitive
Policy is configurable per-tenant. Organizations define their own thresholds and rules.
For Investors & Funders
Zero direct competition at the intersection of cultural protection, FERPA compliance, and re-identification defense.
AI Governance Market
$0.00B
by 2030 · from $940M
AI in Education Market
$0B
by 2030 · from $6.9B
Direct Competitors
0
at this intersection
4 Novel Patent Candidates
Intent × Topic Cultural Classification
Cross-referencing extraction intent categories against protected topic categories for AI prompt classification.
Population-Scaled Re-identification
Community-size multiplier applied to attribute-combination scoring for small-population defense.
Two-Factor Education Record Detection
Requiring co-occurrence of protected data indicators AND student identifiers before flagging.
Cross-Domain Signal Aggregation
Compounding risk multiplier when signals span multiple detection domains simultaneously.
218+
Tests passing
5
Detection domains
<50ms
Scan latency
0
External dependencies
Interested in funding or partnership?
Let's talkGet Early Access
Be the first to know when IndigiArmor launches. Priority access for tribal nations and education organizations.
Our Partners
Building Safer AI Together
We work with organizations committed to protecting sensitive data and empowering communities through technology.
