Entra ID authentication
OIDC sign-in, identity-provider MFA, and role claims protect workforce and tenant access to the control plane.
ActiveSecurity, privacy, and responsible AI
Security and privacy in the private beta.
How the private beta handles identity, authorization, encryption, monitoring, data retention, incident response, and AI use.
Control inventory
What’s active now, and what’s planned.
Every control below is marked active or planned.
Tenant-scoped RBAC
Workspace roles and server-side tenant checks constrain configuration, routing, audit export, and privileged administration.
ActiveManaged identities and Key Vault
Azure managed identities replace embedded service credentials; secrets and certificates are stored in Key Vault and access is logged.
ActiveProtected in transit and at rest
TLS protects public and service connections. Azure-managed encryption protects Cosmos DB, Blob Storage, and Service Bus data at rest.
ActiveTenant-aware data boundaries
Tenant identifiers, partition keys, authorization checks, and scoped audit exports separate each workspace in the shared service.
ActiveWAF, quotas, and rate limits
Front Door WAF policies and API Management limits protect the public API boundary from malformed, abusive, or unbounded requests.
ActiveCentral operational telemetry
Application Insights and Log Analytics collect authentication, policy, latency, failure, and administrative events with alert routing.
ActiveBounded agent authority
Every transaction carries a scope, an expiration, and a spend ceiling. Sensitive routes can require a second authorization before work leaves the boundary.
ActivePre-egress enforcement
Policy-as-code rules are evaluated before anything crosses the boundary, and every allow or deny decision is recorded with the transaction.
ActiveProof without raw data
Receipts carry hashes, pointers, and bounded disclosures instead of raw artifacts by default, with redaction and short-retention options per workspace.
ActiveNo custody of customer funds
Value stays in accounts the customer controls. Where a budget rule requires co-authorization, the policy engine holds an additional key and can never move value alone.
ActiveSecondary-region runbook
A warm recovery environment, replicated configuration, documented objectives, and rehearsed regional failover remain future work.
PlannedDedicated tenant accounts
Separate data accounts and throughput boundaries for customers with stronger isolation requirements remain on the roadmap.
PlannedData handling
What we store, why, and for how long.
The gateway persists configuration and operational metadata while treating routed payload content as transient by default.
| Data class | Handling | Retention |
|---|---|---|
| Account and tenant records | Used for authentication, role assignment, workspace ownership, and support. | Account life plus 30 days. |
| Endpoint and policy configuration | Stored in tenant partitions; sensitive connector values reference Key Vault secrets. | Until deletion plus a 30-day recovery window. |
| Transaction metadata and receipts | The transaction ID, route, timing, outcome class, policy decision, and receipt reference are retained for each transaction; the payload itself is not. | 30 days standard, configurable to 7 days. |
| Customer payload content | Processed in memory for authorized routing and not used to train a shared model. | Not persisted by default. |
| Audit export packages | Encrypted customer-requested packages stored with tenant-scoped access. | 90 days, then lifecycle deletion. |
| Security telemetry | Authentication, administrative, error, and alert events used to detect and investigate misuse. | 90 days in the private beta. |
AI-use boundary
Your team stays in charge of every AI decision.
CGP Gateway may route a customer-authorized request to an AI service or agent. The product does not train a shared foundation model from customer payloads and does not independently approve high-impact outcomes.
Customers choose destinations, configure route policies, validate outputs, and remain responsible for lawful and appropriate use of connected systems.
Customer authorized
Only endpoints registered and allowed by the workspace policy can receive a routed request.
No shared training use
Routed payloads are not used to train a common model or build advertising profiles.
No autonomous high-impact approval
The gateway does not make legal, financial, employment, medical, safety-critical, or eligibility decisions.
Configured by operators
Customer teams define policies, test routes, inspect events, and decide how downstream AI outputs may be used. Consent stays with the principal.
Delegated or autonomous
Audit records distinguish actions a person delegated from actions an agent took on its own initiative.
Incident readiness
How we detect, contain, and recover.
The Chief Cloud and Security Officer coordinates triage, engineering containment, evidence preservation, and recovery, plus customer communication when it's required.
Internal response targets support private-beta operations; they are not a public service-level agreement.
Alert and classify
Centralized alerts and reports enter a severity matrix; critical alerts are acknowledged within a 30-minute target.
Limit access and impact
Operators can revoke identities, rotate secrets, block routes, isolate revisions, and preserve relevant audit evidence.
Restore and review
Runbooks guide validated restoration, post-incident review, corrective actions, and notification where law or contract requires it.
Backups: daily configuration backups, lifecycle-managed encrypted artifacts, and quarterly restore tests are in place. Multi-region recovery and a formally tested regional failover are planned.
Independent assurance
Where we stand on certifications.
IOAI Gateway, Inc. does not currently hold a SOC 2 report, ISO/IEC 27001 certification, PCI DSS attestation, HIPAA assessment, or FedRAMP authorization.
A readiness assessment, control-evidence program, penetration test, and vendor-risk process are planned before any public assurance statement.
Documented beta controls
Policies, implementation records, vulnerability scans, restore tests, and incident exercises are documented internally.
External validation
An independent penetration test and SOC 2 readiness review will precede any public assurance claim.
Precise status only
We publish certification status only when dated evidence supports it.
Security contact
Report a vulnerability.
Use the Contact page and select the security inquiry type, or email us directly using the address below.