Security posture / customer surface

What Island Intel JA protects, and how.

Customer accounts are scoped, revocable, and observable from inside the account. The platform never exposes direct infrastructure access, privileged credentials, or private infrastructure identifiers to a customer session.

Control principles

Customer access stays scoped and revocable.

security

No secrets in the customer surface

No privileged backend credentials or private infrastructure identifiers appear in any customer-facing page or response.

security

Scoped credentials, not raw access

Each customer account receives credentials scoped to the customer-safe tool surface. Scope is bound by the plan, entitlement policy, and organisation.

security

Show-once token rule

MCP credentials issued from inside the customer account are shown once at creation. Only a one-way hash is stored. Tokens are revocable at any time.

security

Revocable agent connections

OAuth 2.1 connections to AI agent hosts can be revoked from inside the customer account. Revocation is enforced before the next tool call.

security

Usage limits enforced server-side

Plan-level rate limits and per-tool-family quotas are enforced by the Island Intel JA gateway, not by the identity provider.

security

Append-only telemetry

Every customer tool call, allowed or denied, is recorded in an append-only telemetry stream the customer can review from inside the account.

security

No privileged credential exposure

Privileged credentials are never exposed to customer sessions, AI agent hosts, or any customer-facing surface.

security

Customer-safe MCP boundary

The MCP tool surface visible to a customer's AI agent is restricted to the customer-safe profile. Operations tools and write paths are not reachable.

security

Documented privacy posture

Telemetry records what tool ran, who called it, and which plan applied. It does not store raw client IP addresses or unnecessary PII.

Responsible disclosure

Report privately before public testing.

Security researchers can report potential issues through the contact path provided to approved accounts. Do not publish or test against production resources without explicit authorisation.