Protopia AI is at RSAC. Meet our team of experts in AI Data Privacy and Security.

Defense and Intelligence

Inference Privacy
for Mission-Critical AI

Sensitive data powers AI that drives outcomes. Protopia Stained Glass keeps it private, so federal agencies can deploy AI faster and maximize infrastructure investments.

status quo

Three paths to private multi-tenant AI

Relying on dedicated infrastructure to serve each sensitive use case guarantees data separation. It also guarantees underutilization and costs that scale with programs instead of users.

Accept the Risk
Codebases, intel products, and operational data co-mingle in plaintext form on model hosting infrastructure. Every tenant’s data visible to the infrastructure and anyone with access. One compromised credential exposes everything.
Isolate at 5–10x Cost
Separate GPU infrastructure per program to prevent co-mingling. Isolated environments run at 20-40% utilization, driving costs 5-10x higher. Finer-grained isolation increases the multiplier further.
Transform data, share GPUs
Stained Glass Transform protects data before it reaches the model hosting infrastructure. Models never see plain text. One shared GPU cluster at 90%+ utilization.
See the economics →
Architecture

Inference privacy on any infrastructure

Stained Glass converts prompts and context into protected representations that are processed by AI models instead of plaintext.

Federal Agency
Enterprise Zone of Trust
PROMPT
Determine benefits eligibility for this service member
CONTEXT
DD214.pdf
DATA SOURCE
Personnel Files
Stained Glass Transform
protected
embeddings
response
Model Hosting Infrastructure
No plaintext present
FROM FEDERAL AGENCY
[0.42, -0.18, 0.91, 0.07, -0.55…]
no plaintext
MODEL ENDPOINTS
NVIDIA Nemotron
INFRASTRUCTURE
vLLM NVIDIA NIMs HPE
  • Plaintext stays in the tenant boundary
    The inference infrastructure never processes plaintext prompts or context.
  • Works with your existing stack
    Open AI API-compatible. No model retraining. No specialized hardware.
  • Accuracy parity with plaintext
    Validated across HellaSwag, MMLU, TruthfulQA, and ARC.
Deployment Patterns

Unlock AI use Cases for Defense and Intelligence

Validated across classified, hybrid, and edge environments.

Cross-Agency Collaboration
Multi-Tenant Inference
Enable multiple agencies, commands, or mission partners to safely run inference on the same GPUs. Models operate on transformed data, eliminating exposure in plaintext to any tenants, platform operators, or unauthorized systems.
Tactical Edge
Overflow from Edge to Cloud
Surge inference workloads from constrained edge devices to more capable models hosted in the cloud. Stained Glass transforms sensitive data before it leaves the local boundary, so you scale compute freely across classification levels.
MULTI-SOURCE INTELLIGENCE
Private AI Across Data Sources
Process multi-INT and multi-sensor data in inference pipelines without data co-mingling. Stained Glass protects source-identifying features in the prompt, enabling analysts to query shared models across data types and compartments.
U.S. Navy Case Study

Private Computer Vision for VIP Detection

SGT enables computer vision models to detect key individuals from video feeds without exposing raw footage.

Benefits

Protopia AI delivers broad value to DoW

Lower TCO
Leverage efficient shared AI infrastructure across teams and agencies to reduce cost and increase GPU utilization.
Retain Data Ownership
Reduce risk and align with federal Zero Trust mandates by keeping sensitive data within the agency’s trust zone.
Accelerate Mission Impact
Unlock previously restricted data (e.g. video, image, text, code) for defense, intelligence, and civilian use cases.
Proof points

AI that meets DoW standards for trust and scale

One goal is to find ways to deliver AI capabilities that can serve 80 percent of the Army…these efforts are hyper-focused on back office, and on capabilities that are ready to go.

LG
Leonel Garciga
US Army Chief Information Officer

The Navy is accelerating drone and AI technology development under field conditions, ensuring that unmanned platforms can operate with greater autonomy and resilience in contested environments.

JFIX
2025 JFIX Report
U.S. Navy
U.S. Army
XTech
1st Place Winner 2024
U.S. Air Force
49th Maintenance Group
$1.25M D2P2 2025
U.S. Army
Enterprise Deplyoment
$2M D2P2 contract Q1 2025.
U.S. Air Force
AFSC/SW
$1.25M D2P2 2025