Security
Iron Forged ARI uses a fail-closed authorization model for protected application data and keeps public AI surfaces intentionally read-only.
Implemented source controls
- Protected customer access is authenticated server-side and resolved through authoritative
UserAuthorizationstate. - Customer identifiers for protected reads are derived server-side rather than accepted from client-supplied scope.
- Customer-facing records are projected through explicit field allowlists; schema additions are not automatically exposed.
- Public MCP and A2A source surfaces expose no outreach-send, Qualification mutation, Revenue Judgement mutation, booking, checkout, or commercial-commitment operation.
- External AI customer access is designed to require both approved authentication and a verified, active entitlement for the requested right.
- AI distribution endpoints include bounded request IDs, minimal operational logging, and an application-instance rate-limit guard. This guard is defense-in-depth and is not represented as a distributed WAF or global rate-limit service.
Secrets and reporting
Do not send passwords, tokens, API keys, or authentication secrets by email. Security reports may be sent to support@ironforgedari.com. See security.txt.
Assurance boundary
Source controls are not Production proof by themselves. Runtime deployment, external platform configuration, penetration testing, certifications, and independent audits must be evidenced separately before they are claimed.