The Execution &
Evidence Layer for AI

We run your AI safety controls and prove they executed. Third-party witnessed, zero data egress.

● Deploy Faster

● Scale Without HITL

● Unlock High-Stakes Use Cases

The Problem

AI deployment is blocked

You built the AI. It works. But you can't ship it—because you can't prove it's safe.

🚫

Compliance says no

"Show us your audit trail." You have logs, but no proof your safety controls actually ran.

⏳

Scale requires reviewers

Human-in-the-loop for every decision. Your unit economics collapse as volume grows.

🔒

High-stakes locked out

Healthcare, finance, legal—where AI creates the most value, trust gaps block deployment.

The gap: You need to prove your AI is trustworthy—but existing tools only document what happened. They don't execute controls or prove execution.

The Solution

We run your safety controls
and prove they executed

GLACIS is a control plane that sits alongside your AI stack—executing policies, not just logging them.

What GLACIS Does

1. Runs your safety controls in-line with inference
2. Attests every execution with cryptographic proof
3. Witnesses via third-party transparency log
4. Zero egress—your data never leaves your infra

Control Plane, Not Competitor

We don't replace your guardrails, evals, or safety tools. We orchestrate them—and prove they ran.

Guardrails ✓ Evals ✓ LLM Firewalls ✓ Your tools ✓

The harmonizing layer that makes your existing stack auditable.

The result: Proof that your AI safety controls executed—verifiable by regulators, auditors, and customers.

What This Unlocks

Deploy faster. Scale bigger. Unlock more.

🚀

Deploy in weeks, not months

Compliance pre-solved. Ship AI features while competitors are still in security review.

📈

Scale without adding reviewers

Automated execution + proof replaces human-in-the-loop. Unit economics that actually work.

💎

Unlock high-stakes use cases

Healthcare diagnostics. Financial decisions. Legal review. Where AI creates the most value.

The bottom line: Trustable AI = deployable AI = more revenue.

How It Works

Sidecar architecture. Zero egress. Third-party witnessed.

Your Infrastructure

AI Application

+ Guardrails, evals, etc.

→

GLACIS Sidecar

Execute + Attest

Runs in YOUR VPC

→

GLACIS Network

Third-Party Witness

Only hashes transmitted

pip install glacis

5-minute integration

Zero data egress

PHI/PII stays local

Auditor-verifiable

Cryptographic proof

Traction

Strong pull. Just getting started.

Next milestone

JPM Healthcare 2026

Design Partner

nVoq • Ambient AI for home care

50k+

visits/month

Colorado-based. Facing Colorado AI Act deadline.
Consent attestation • PHI proof • Guardrail evidence

In Pipeline — Pharma

Praxis Pro • Pharma sales training AI

The signal: Tenant-bounded AI with trade secrets, pre-release drug data, competitive intel. Need to prove what went in and came back from every query.

Also in pipeline

Prompt Opinion

+ 3 more

Inbound interest

Health Systems

AI governance inquiries

AI Vendors

Deployment friction resonates

Investors

Regulated AI + governance thesis

Security Benchmark

Attack Success Rate by Category

Healthcare AI security benchmark across 6 frontier models

RAW (No Protection)

With GLACIS

Consent Fabrication

44.4% 0%

Agentic Exploits

7.4% 0%

Multi-turn Attacks

6.2% 0%

Documentation Integrity

5.6% 0%

Guardrail Degradation

2.5% 2.4%

1,776 tests across 6 frontier models · 0% false positive rate on 1,200 benign requests

Model Analysis

Every Frontier Model is Vulnerable

The problem isn't one model—it's systemic

GPT-4o

OpenAI

7.6%

RAW

1.0%

GLACIS

Gemini 3 Pro

Google

5.2%

RAW

0%

GLACIS

Gemini 3 Flash

Google

5.2%

RAW

2.1%

GLACIS

Claude Sonnet 4.5

Anthropic

4.2%

RAW

0%

GLACIS

Claude Opus 4.5

Anthropic

2.8%

RAW

2.1%

GLACIS

GPT-5.2

OpenAI

2.1%

RAW

2.1%

GLACIS

0% false positive rate across all models — 1,200 benign requests tested

Team

FDA Authorized. Enterprise Deployed.
We've lived this problem.

Joe Braidwood

Co-Founder & CEO

SwiftKey → 1 in 4 smartphone users

Founding exec, $250M Microsoft exit. Chief Strategy at Vektor Medical—secured reimbursement for AI device. Cambridge Law.

Dr. Jennifer Shannon

Co-Founder & CMO

Cognoa → First FDA De Novo for AI diagnostics

Medical Director at Cognoa. Navigated FDA authorization for AI that diagnoses autism in children.

The Ask

$2M to make AI deployment trustable

Use of Funds

Engineering 50%

2 senior engineers (Rust, infra)

Go-to-Market 30%

Enterprise sales, design partners

Operations 20%

Legal, compliance, infrastructure

18-Month Milestones

1

5+ production deployments

Execution + evidence in production

2

$500K ARR

Platform + enterprise contracts

3

Category definition

Execution & Evidence as industry standard

4

Series A ready

Metrics for $8-12M raise

The Execution & Evidence Layer—making AI trustable, deployable, and scalable.

Thank you

The Execution & Evidence Layer for AI

joe@glacis.io
glacis.io

The Execution &Evidence Layer for AI

AI deployment is blocked

Compliance says no

Scale requires reviewers

High-stakes locked out

We run your safety controlsand prove they executed

Deploy faster. Scale bigger. Unlock more.

Deploy in weeks, not months

Scale without adding reviewers

Unlock high-stakes use cases

Sidecar architecture. Zero egress. Third-party witnessed.

Strong pull. Just getting started.

Attack Success Rate by Category

Every Frontier Model is Vulnerable

FDA Authorized. Enterprise Deployed.We've lived this problem.

$2M to make AI deployment trustable

Thank you

The Execution &
Evidence Layer for AI

We run your safety controls
and prove they executed

FDA Authorized. Enterprise Deployed.
We've lived this problem.