So here's the page that respects that. Five questions every security review of an AI platform ends up asking, answered directly — with links to where each claim can be inspected rather than taken on faith. If your questionnaire asks something this page doesn't, send it to hello@turtleaicoworker.com and we'll complete it.
The pattern across every answer: controls live in the execution path, not in a policy PDF. Isolation is server-side, enforcement is per tool call, and evidence is a lookup.
Every object — table, agent, credential, policy, audit row — carries its organization and workspace scope, and every query is filtered server-side. Isolation is a data-layer property, not a UI convention. Access follows one role × action matrix (Owner / Admin / Builder / Member), identical across web, mobile and API. Credentials are encrypted at rest and decrypted only at request time; the API returns them masked, never in the clear.
The full posture on /trustExactly what its autonomy scope and policies allow — enforced at runtime, not requested in a prompt. The policy engine evaluates every tool call before it executes, and a runtime authorization floor makes unapproved writes a no-op, even if a prompt injection put them in the plan. Approvals are single-use and expire in 24 hours; the next matching call asks again.
Policy engine mechanicsWorkspace content never trains models, ours or anyone else's. Only the provider an agent is configured to use ever sees a prompt, and with BYO keys those requests run under your own provider agreement. A prompt-injection shield screens untrusted content — tool results, knowledge chunks, table rows, webhook payloads — before it re-enters a model's context.
Data handling commitmentsEvery run and every tool call is recorded: integration, function, read-vs-write classification, sanitized inputs and outputs, PII flags, token and dollar cost. Secrets are redacted before storage. Runs pin the immutable config version they executed under, so “what was this agent allowed to do in March” has one exact answer. Filterable, exportable, with per-workspace retention and legal hold.
Audit and analyticsCircuit breakers auto-pause a worker when a cost or error threshold trips. Hard daily and monthly budget caps stop spend at the line, with breach-date forecasting before you get there. Kill switches halt a worker outright, and every governance event — the block, the pause, the override — lands in the same audit trail as everything else.
Approvals and safeguardsForward these to whoever owns the assessment. Each stands on its own.
Tenancy isolation, identity, runtime enforcement, framework mapping — the whole posture on one page, written for a reviewer.
Read /trustThe policy engine in detail: rule types, verdicts, and where enforcement sits in the execution path.
See the mechanicsA short self-assessment of your current AI controls — useful for framing what to inspect here.
Run the checkArchitecture overview, data-flow diagrams, compliance status and completed questionnaire answers, on request.
Request by emailAlso useful: the readiness guide for rolling out safely, and pricing — enterprise controls (SSO, audit exports, data residency, self-hosting) live on the Enterprise plan.
Security questionnaire, architecture questions, a call with someone who can actually answer them — answering security reviews is part of the job, not a favor.
