Security, privacy, and data

Atoi’s product safety model combines scoped authorization, explicit human gates, inspectable receipts, and revocable access. This page describes product behavior, not a legal warranty. The current Privacy Policy and Terms remain authoritative for legal commitments.

#Credentials

Provider keys and authorization tokens are secret inputs. Atoi masks stored credentials, keeps server-owned provider secrets away from client-visible records, and does not ask an agent to type payment, passkey, multi-factor, or other high-risk credentials.

Never paste a key into a conversation, document, screenshot, support report, or public share.

#Permissions

Connections and computers use explicit scope. Routine read-only work can proceed under the current policy; outward-facing, destructive, credential-sensitive, payment, permission, and settings actions retain a human gate.

Read Asks, approvals, and safety for the user-facing model.

#Sharing

Private participation and public publication are separate. Review conversation history before inviting a person. Review selected content and redaction information before publishing. Revoke the exact participant, invitation, or public share whose access should end.

#Execution data

Tasks can produce logs, diffs, screenshots, recordings, and artifacts. Treat them according to the source data they contain. A proof artifact is useful evidence, but it can also contain paths, names, or account state that should not be published.

#Account data

Supported account controls can expose export or deletion workflows. Read each confirmation carefully: removing an account record, stored attachment, external authorization, or public snapshot may have different consequences.

#Report a security concern

Do not include live secrets. Record the affected surface, minimum reproduction, observed access boundary, expected boundary, and whether the behavior is current. Use the official security contact or reporting path linked from Atoi’s public legal surface.