Clear controls. Honest boundaries.
Security should be specific. Here is what the current pilot protects and what still requires implementation.
Identity and private records
Private workspaces require ChatGPT sign-in. Application routes scope records to the authenticated account. The customer inquiry inbox is separately restricted to the configured Prodigi owner. This is an individual-workspace model, not shared company membership or enterprise SSO.
Encryption and access
Event, portfolio, decision and inquiry payloads use AES-256-GCM encryption. Evidence file contents are encrypted before object storage. Source keys are stored as hashes and support rotation and revocation. Operational metadata, including identifiers and timestamps, is not application-encrypted. The server can decrypt records; this is not end-to-end encryption.
Application controls
Validation, origin checks, CSRF protection for signed-in changes, request-size limits and basic rate limits protect supported workflows. Trial requests use signed form tickets and separate consent choices. Uploaded evidence is not malware-scanned and is downloaded rather than rendered inline.
Execution boundary
Applications must call and honor the decision API. Pausing a source blocks its Prodigi ingestion and new permits; it does not shut down a machine. Models and recorded authority profiles do not prove that external permissions changed.
Assurance
The platform is a pilot. No independent penetration test, compliance certification, guaranteed availability or automatic remediation claim is made. Use selected non-sensitive test data while evaluating it.
