Sentinel Core +
Sentinel Guard

AI interaction audit custody — built for OSFI E-23. Guard surfaces what your employees are doing with AI. Core proves it.

◈ Sentinel Core ◆ Sentinel Guard

Sentinel Core Overview v3.0 — Technical Briefing — April 2026

The Regulatory Obligation

OSFI E-23 takes effect May 1, 2027

Every federally regulated financial institution must produce runtime evidence of AI model lifecycle governance — not a policy document, not a risk framework. Evidence.

1

Recorded at the time it occurs

Every AI interaction captured the moment it happens.

2

Tamper-evident

Unalterable after the fact. Cryptographically chained, Merkle-anchored, externally timestamped.

3

Independently verifiable

On demand by an examiner, auditor, or board — no OAIS involvement required.

Two Products. One Audit Chain.

Guard surfaces it. Core proves it.

◈ Sentinel Core

SDK Wrapper — Developer Integration

Every programmatic AI API call. Supports Anthropic, OpenAI, Azure OpenAI, Gemini, Bedrock. No network changes. No CA installation. No admin rights.

◆ Sentinel Guard

Browser Extension — IT Deployment

Every web-based AI interaction — ChatGPT, Claude.ai, Gemini, Copilot. Deployed via Intune or Chrome Enterprise Policy. No code change. No user action.

Together: a single verifiable audit record covering both your systems and your employees — including shadow AI.

◈ Sentinel Core

What Core does on its own

Core is a standalone SDK wrapper. No browser extension. No IT deployment. One integration — full AI audit custody.

🔍 Capture Every AI Call

Wraps your existing AI client — Anthropic, OpenAI, Azure OpenAI, Gemini, Bedrock. Every prompt, response, and token count captured at execution.

🔒 Tamper-Evident Audit Chain

HMAC-hashed, Ed25519-signed, Merkle-anchored, RFC 3161 timestamped. Immutable. Independently verifiable. No trust in OAIS required.

📈 Model Intelligence

Track which models your systems use, token spend across providers. Optimize model selection, detect drift — all from captured interaction data.

Who needs Core alone? Teams running AI in production pipelines, agents, or backend systems — where there is no browser to monitor.

IT Admin View

Two deployment paths. Same audit chain.

◆ Sentinel Guard Enterprise / MDM

⚙

1. Admin configures policy

Set DLP rules, AI provider restrictions, and escalation workflows in the Guard dashboard.

📤

2. IT pushes extension silently

Deploy via Microsoft Intune or Chrome Enterprise Policy. No user action required.

✅

3. Employees see nothing change

Guard icon appears in browser toolbar. AI interactions are monitored and governed automatically. Zero end-user friction.

◈ Sentinel Core Developer / CLI

ai-engine — onboarding

◈ Core Captures & proves every AI interaction

+

◆ Guard Adds DLP enforcement & browser policy

Each works standalone. Together they cover every AI interaction surface.

◆ Sentinel Guard

DLP Dashboard — Configure. Enforce. Audit.

0

Total Rules

0

Block

0

Escalate

0

Active

User	Email	Role	Department	Status	Actions
No users configured — click + Add User to get started

User	Rule Pattern	Reason	Expires	Actions
No active exemptions

◆ Sentinel Guard

Extension activation — Connect. Enforce. Protect.

🔒 claude.ai

In managed environments, administrators push the extension silently via Intune or Chrome Enterprise Policy — users never need to authenticate manually. AI provider access can be blocked entirely unless Guard is active.

◆ Sentinel Guard

Real-time DLP — Block. Hash. Audit.

🔒 claude.ai/chat

△ Guard Active

Claude

How can I help you today?

↑

🛡

Sentinel Guard — Data Blocked

PCI DSS: Credit Card Number Detected

            4532 8921 0044 7891
            ↓ SHA-256 Hash
            9f86d081884c7d659a2feaa0c55ad015a3bf4f1b2b0b822cd15d6c15b0f00a08
          

Message blocked per DLP policy. Card number hashed for audit trail.

✖ Block & Log

↑ Elevate for Approval

        [14:32:07]
         BLOCKED PCI_CREDIT_CARD in claude.ai —
        SHA-256: 9f86d08... —
        Policy: BLOCK+HASH | Elevate: available
      

Data Sovereignty

What you see vs. what OAIS receives

💚 Your System (Client Environment)

RAW CONTENT — STAYS IN YOUR ENVIRONMENT

INPUT:
"Recommend top 3 extended warranty products for 2024 Honda CR-V, customer profile: young family, 60k km/yr, budget tier mid..."

OUTPUT:
"Based on the profile, I recommend: 1) Premium Shield Plus (comprehensive, $1,840/4yr), 2) DriveGuard (powertrain, $1,290/4yr), 3) Essential..."

HMAC
SHA3-256

hashes
only

🔒 OAIS Registry (What We Receive)

IRREVERSIBLE HASHES ONLY

input_hash:
waiting...

output_hash:
waiting...

model_id: claude-sonnet-4-6
gate_decision: ALLOW
signature: waiting...

◆ Sentinel Guard

Browser-level enforcement. Real-time.

🔒 claude.ai/chat

△ Guard Active

Employee using Claude for claims analysis

You

Review claim #4419: policyholder Jane Doe, SIN 904-271-638, policy EW-2024-88412...

BLOCKED Sentinel Guard — Sensitive Data Detected

● Social Insurance Number detected (904-***-***)
● Policy number pattern detected
Prompt blocked before leaving browser. Metadata record transmitted to Sentinel chain. Raw content never leaves the endpoint.

Chrome & Edge. Deployed via Intune or Chrome Enterprise Policy. Captures AI tools adopted without IT approval.

Sentinel Dashboard

Live interaction monitoring

0

Total Interactions

0

Allowed

0

Flagged

0

Blocked

Interactions — Last 7 Days

Recent Interactions

Audit Readiness

OSFI examiner requests evidence. You're ready.

📩

Request Received

OSFI examiner requests AI interaction evidence

📦

Export Bundle

Records, Merkle proofs, TSA tokens packaged

🔎

Chain Verification

verify_chain.py — no OAIS dependency

🕒

TSA Validation

Sectigo RFC 3161 timestamps confirmed

✅

Verdict

VERIFIED — independently, permanently

OSFI examiner workstation — no OAIS access required

Coverage

What Sentinel captures today — and next

Surface	Status	Coverage
Programmatic AI API calls	LIVE	Python systems, agents, pipelines — Anthropic, OpenAI, Azure OpenAI, Gemini, Bedrock
Browser-based AI tools	LIVE	All web-based AI in Chrome/Edge — including shadow AI
M365 Copilot (Purview)	Q2 2026	Purview audit logs via Graph API. Word, Excel, Teams, Outlook, PowerPoint. ~2 weeks to deploy.
VS Code Extension	Q2 2026	GitHub Copilot, Cursor, Codeium — AI code suggestions
Additional SDK Wrappers	ROADMAP	Node.js, Java, .NET

Data Sovereignty

Your content never leaves. Only hashes do.

Your content is hashed using a secret that lives in your infrastructure. OAIS never holds this key, never transmits it, and cannot retrieve it. This applies to both Core and Guard — regardless of which product you deploy.

🔒 What OAIS holds

✓ Cryptographic hash values
✓ Ed25519 public keys
✓ Merkle proofs
✓ RFC 3161 TSA tokens

🚫 What OAIS never holds

✗ Your HMAC secret key
✗ Your raw AI inputs or outputs
✗ Any reversible content data
✗ Your AI interaction history

The system is live. Let's start.

709

Real AI interactions
in verified chain

287

Automated tests
7 build phases

0

Chain gaps
since deployment

1

Technical walkthrough with your engineering team
Review AI systems, confirm SDK compatibility, plan Guard rollout.

2

Pilot deployment — Core + Guard on live traffic
Hashes flowing. Merkle anchoring. Guard capturing browser usage.

3

Proof of concept — tamper-evident audit record
Independently verifiable. OSFI-ready. Permanently anchored.

Business continuity: Third-party escrow. Full export within 30 days of any cessation event. RFC 3161 tokens verify independently and permanently.

info@oais.ai | OAIS.ai | Ontario, Canada

Choosing Your Path

Which product is right for you?

All options share the same data sovereignty model — your content never leaves your environment.

◈ Sentinel Core

Core Only

For teams running AI in production pipelines, agents, or backend systems.

✓ Full audit custody of programmatic AI calls
✓ Model intelligence & token spend analytics
✓ Tamper-evident chain (HMAC + Ed25519 + Merkle)
✓ Drift detection & optimization insights

◈ Core ◆ Guard

Core + Guard COMPLETE COVERAGE

Cover every AI interaction surface — programmatic and browser-based — in one audit chain.

✓ Everything in Core
✓ Browser-level DLP enforcement
✓ Shadow AI detection & governance
✓ Silent MDM deployment (Intune / Chrome Policy)

◆ Sentinel Guard

Guard Only

Browser AI control is your primary concern — block, flag, or escalate at the browser level.

✓ Real-time DLP for browser-based AI tools
✓ Shadow AI visibility & blocking
✓ Silent enterprise deployment
✓ Audit chain for all browser AI interactions

Data sovereignty across all options — your HMAC key stays in your infrastructure. O.A.I.S. receives only irreversible hashes.

Sentinel Core +Sentinel Guard

The Regulatory Obligation

OSFI E-23 takes effect May 1, 2027

Recorded at the time it occurs

Tamper-evident

Independently verifiable

Two Products. One Audit Chain.

Guard surfaces it. Core proves it.

SDK Wrapper — Developer Integration

Browser Extension — IT Deployment

What Core does on its own

🔍 Capture Every AI Call

🔒 Tamper-Evident Audit Chain

📈 Model Intelligence

IT Admin View

Two deployment paths. Same audit chain.

DLP Dashboard — Configure. Enforce. Audit.

Generate Rules from Policy Document

Edit Rule

Add New User

Grant Exemption

Extension activation — Connect. Enforce. Protect.

Real-time DLP — Block. Hash. Audit.

Data Sovereignty

What you see vs. what OAIS receives

💚 Your System (Client Environment)

🔒 OAIS Registry (What We Receive)

Browser-level enforcement. Real-time.

Sentinel Dashboard

Live interaction monitoring

Audit Readiness

OSFI examiner requests evidence. You're ready.

Request Received

Export Bundle

Chain Verification

TSA Validation

Verdict

Coverage

What Sentinel captures today — and next

Data Sovereignty

Your content never leaves. Only hashes do.

🔒 What OAIS holds

🚫 What OAIS never holds

The system is live. Let's start.

Choosing Your Path

Which product is right for you?

Core Only

Core + Guard COMPLETE COVERAGE

Guard Only

Sentinel Core +
Sentinel Guard