Trust Center
How AISOS earns trust.
These are the security practices implemented in the product today. Our controls are mapped to SOC 2 and ISO 27001, and the console produces signed compliance exports designed to support those audits.
Policy boundary on every call
Every prompt, document, tool call and answer is checked against your organisation's policy before any model sees it. Admission is deny-by-default.
Tamper-evident records
Administrative actions and decisions are written to a hash-chained ledger. Organisations can publish signed fingerprints, and anyone can verify them on the Verify page — no account needed.
Tenant isolation
Each organisation's data is locked to its members and enforced at the database level, not just in the application.
Strong authentication
Two-factor authentication with authenticator apps, organisation-level enforcement of it, and company single sign-on (SAML) support.
Scoped, rate-limited API keys
Keys are stored hashed, carry scopes, optional expiry and per-minute rate limits, and record last use.
Memory safety
Agent memories carry trust labels; poisoned content is quarantined, and revoking a source revokes everything derived from it.
Payment security
Card payments are processed by Paddle; card numbers never touch AISOS servers. Webhook events are signature-verified and applied once.
Data residency & retention
Data is hosted on managed cloud infrastructure (database, authentication and file storage) with model inference routed to the upstream providers listed on the Subprocessors page. Retention is configurable per organisation by its owner:
| Data | Default | Configurable range |
|---|---|---|
| Usage events (prompt metadata, costs, decisions) | 365 days | 30–3,650 days, owner-set |
| Admin record (hash-chained ledger) | Kept indefinitely | Never purged — by design |
| Account & organisation data | Until account deletion | — |
The same retention rules apply on every plan; there is no zero-retention option today. Purging usage data never touches the admin record.
Audit-log export
Organisation owners, admins and auditors can export the full admin record as JSON or CSV from the console's Audit Trail page, with a chain-verification result included so the export can be checked offline. Signed ledger fingerprints can be published and verified publicly on the Verify page.
Subprocessors & questionnaires
Our subprocessor list is updated whenever a provider changes — check the Subprocessors page for the current list. For security questionnaires (SIG, CAIQ, custom), contact us through the contact page with "Security questionnaire" in the message.
Report a vulnerability
Security researchers: report vulnerabilities through the contact page with "Security" in the message. Our disclosure policy is published at /.well-known/security.txt.