Trust center · Completed questionnaire

Your security questionnaire,
answered before you send it.

33 answers across nine domains — the ground a CAIQ or SIG review covers, in plain language, published rather than gated behind a sales call. Every answer is verified against the codebase, most recently July 2026. Need it in your own questionnaire format? Send it to hello@turtleaicoworker.com and we'll complete it.

Read the full trust center
01

Organization & governance

3 items
Formal security programYesPlatform-enforced controls — policy engine, audit, tenancy — with an active security program; ISMS documentation being formalized alongside SOC 2 preparation.
CertificationsYesSOC 2 Type II in progress with an independent auditor; the report will be published on the trust center when complete. Nothing is claimed before it's earned.
Security contactYessecurity@turtleaicoworker.com — monitored, and answering questionnaires is part of the job.
02

Identity & access

6 items
Single sign-onYesGoogle and Microsoft OAuth2 — access follows your identity provider, offboarding follows your directory.
Multi-factor authenticationYesTOTP two-factor with hashed backup codes, enforced at login when enabled.
SessionsYesHard cap of 4 concurrent sessions per user; sessions are enumerable and revocable.
TokensYesShort-lived access tokens over rotating refresh tokens with blacklist-after-rotation — a stolen refresh token dies on first reuse.
Role-based accessYesFour roles — Owner, Admin, Builder, Member — through one role × action matrix, identical across web, mobile and API.
PasswordsYesPBKDF2 hashing with length, similarity, common-password and numeric-only validators.
03

Application & API security

5 items
Rate limitingYesPer-scope throttling on anonymous and authenticated traffic.
Injection defensesYesORM-parameterized queries throughout; user-supplied sort and filter parameters are resolved against real column definitions and never string-interpolated.
Headers & TLSYesTLS enforced with HSTS (one year, subdomains, preload), nosniff, X-Frame-Options DENY and secure cookies; outbound calls verify certificates.
File uploadsYesType and size validation on upload.
SecretsYesAll secrets environment-injected with no fallbacks — the application refuses to boot without its secret key. Explicit host and CORS allowlists.
04

Data security & encryption

4 items
In transitYesTLS for all client traffic.
At restYesData stores encrypted at rest; integration OAuth tokens and custom-tool credentials additionally application-encrypted and only ever returned masked.
Credentials vs. models & logsYesOAuth tokens never enter model context. Audit records are sanitized before storage: passwords, tokens, keys and card numbers are redacted.
DeletionYesWorkspace and organization deletion cascades to tables, knowledge bases, runs and credentials; derived stores (vectors, files) are reconciled within 24 hours.
05

Data privacy & PII

3 items
PII controlsYesPer-workspace PII policy detecting emails, phones, SSNs, cards, addresses and names — mask, hash, remove or flag, with allowlist patterns.
DPAYesAvailable on request, with standard contractual clauses where applicable.
SubprocessorsYesPublished on the trust center. Model providers are engaged only per your per-agent configuration; BYO keys run under your own provider agreement.
06

Tenancy & isolation

1 item
Customer isolationYesEvery object carries organization and workspace scope, and every query is filtered server-side through a central tenancy layer with enforced org context.
07

Logging, audit & evidence

3 items
AuditabilityYesEvery run and every tool call recorded — integration, function, read/write classification, sanitized inputs and outputs, PII flags, token and dollar cost. Filterable, exportable.
Tamper-evidenceYesImmutable, content-hashed configuration versions; every run is pinned to the version it executed under.
RetentionYesPer-workspace retention policies with legal hold that freezes purging — stamped with who, why and when.
08

AI-specific controls

6 items
TrainingYesYour data never trains models — ours or any provider's. No training code path exists.
Model exposureYesOnly the providers an agent is explicitly configured to use ever see a prompt.
Human oversightYesA policy engine evaluates tool calls before execution across all execution paths. Approvals are single-use grants with expiry; unapproved writes do not execute.
Prompt injectionYesUntrusted content — tool results, knowledge chunks, table rows, webhook payloads — is screened before re-entering model context: flag, wrap or block.
Runaway controlsYesHard daily and monthly budget caps that stop work at the limit, breach-date forecasting, and circuit breakers that auto-pause a worker on cost or error thresholds.
Framework alignmentYesControl evidence mapped to SOC 2, ISO/IEC 42001, NIST AI RMF and the EU AI Act's oversight expectations — a control mapping, not a certification claim.
09

Availability & continuity

2 items
BackupsYesAutomated backups with point-in-time recovery at the database layer.
Incident responseYesDocumented incident-response process with severity levels and customer notification commitments, detailed in the DPA.

Certifications in progress are listed as in progress on the trust center — this page answers control questions, it does not claim certificates. Verified against the codebase: July 2026.

Still have a hard question?

Good. Send it.

Custom questionnaires, architecture walk-throughs, or the DPA — answering your security team is part of the job, not a favor.

Contact security