Colorado AI Act Jun 30, 2026 | EU AI Act Aug 2, 2026 | California ADMT Jan 1, 2026
Zero-Egress Runtime Verification

Run AI inside your boundary.
Prove every control ran.

A lightweight sidecar that lives inside your VPC, enforces configurable controls on every AI inference call, and emits independently verifiable receipts. Plaintext PHI stays in your environment. Whether a BAA is required depends on deployment configuration and your organization’s HIPAA analysis.

Only hashes cross the boundary

The GLACIS sidecar deploys inside your VPC as a Docker container or Kubernetes sidecar. It intercepts AI inference calls, runs your configured controls, and turns control execution into independently verifiable receipts—all designed so plaintext prompts, responses, and PHI never leave your environment.

GLACIS is architecturally incapable of receiving your data

<10ms

Overhead at standard attestation level

80+

PHI detection pattern categories

Any Model

OpenAI, Anthropic, Gemini, open source

Days

Not months. Docker or Kubernetes sidecar.

Every inference call. Every control. Verifiable.

The sidecar runs your configured controls on every request and response, generating receipts that prove each control executed as configured.

PHI Detection

80+ pattern categories for protected health information. Names, MRNs, dates of birth, diagnoses, and dozens more—caught before they reach the model.

Consent Verification

Validates consent status before inference execution. Ensures every AI interaction has the required authorization chain.

Jailbreak Detection

30+ threat patterns for prompt injection and jailbreak attempts. Blocks adversarial inputs before they reach your model.

Configuration Drift

Continuous monitoring of control configuration integrity. Detects and evidences any changes to your control settings.

Unicode Smuggling

Detection of obfuscated content in Unicode and encoded payloads. Catches hidden instructions embedded in seemingly normal text.

Recursive Decoding

Base64 and multi-layer encoding detection and unwinding. Peels back nested encodings to inspect what’s actually being sent.

Stuck in security review? We built this for you.

AI vendors stuck in hospital security review. Your product works. Their security team won’t sign off. GLACIS gives them independently verifiable evidence that controls ran—not just a promise that they will.

Agent developers who need governance infrastructure but don’t have it. You’re building AI agents, not compliance tooling. Embed GLACIS and get the governance layer your customers require without building it yourself.

Digital health companies deploying AI into clinical workflows. When PHI touches AI inference, you need evidence that the right controls ran. Every time. On every call.

Any organization where PHI touches AI inference. If protected health information is anywhere near an AI model, you need zero-egress architecture and evidence to prove it.

You ship agents.
We ship the verification substrate.

Your customers get independently verifiable proof that controls ran. You get through security review. Embed GLACIS into your agent infrastructure and ship with confidence.

Embed

Drop the sidecar into your agent infrastructure. One container, standard API interface.

Evidence

Every inference gets a cryptographic evidence record. Third-party witnessed, independently verifiable.

Ship

Hand your customer an evidence trail. Get through security review. Close the deal.

Live in days, not months

$25–50K

Per year, per deployment environment

Days

Live in your environment—not months of integration work

Common questions about Deploy

What does “zero-egress” mean?

Your data—prompts, responses, patient information—never leaves your VPC. The GLACIS sidecar processes everything locally. Only cryptographic hashes cross the trust boundary to our independent witness for evidence recording.

Do we need a BAA?

The GLACIS sidecar is designed so that plaintext PHI stays in your environment — only cryptographic commitments cross the trust boundary. Whether a BAA is required depends on your specific deployment configuration and your organization’s HIPAA analysis. This architecture is designed to minimize BAA scope, not to bypass it.

How does deployment work?

The GLACIS sidecar deploys as a Docker container or Kubernetes sidecar within your existing infrastructure. Typical deployment takes days, not months.

What model providers are supported?

Any provider that accepts HTTP API calls—OpenAI, Anthropic, Google Gemini, Azure OpenAI, and any open-source model with an API interface.

What’s the performance overhead?

Sub-10ms at standard attestation level. Configurable based on your throughput requirements and evidence depth needs.

Deploy is the runtime layer. Here’s the rest of the stack.

Assess

Know where you stand

3–4 week governance assessment benchmarked against ISO 42001 and NIST AI RMF.

Book an assessment

Comply

Continuous AI governance

The compliance platform purpose-built for AI systems. Multi-framework mapping, evidence generation, OSCAL export.

Request a demo