Data privacy is the first question we get asked, ahead of price and ahead of features. That is the correct instinct: you are considering putting pre-release financial information into a tool that uses AI. This page answers the question directly, and the trust center below has the artifacts.
The short version
SOC 2 Type 2
Zero data retention
Licensed authoritative sources
Not a settings toggle
Why zero data retention is the question that matters
Most AI data-privacy discussion focuses on whether a vendor trains on your data. That matters, but it is only half the question. The other half is what happens at the model layer — the provider actually running the inference.
A tool can have an excellent privacy policy of its own and still pass your content to a model provider that retains it for thirty days, or uses it to improve their service, or both. The commitment has to extend all the way down the chain, and it has to be contractual rather than configurable.
That is why we run zero-data-retention agreements with each provider we route to rather than relying on default API terms. It also means the answer does not change when someone on your team or ours adjusts a setting.
Multi-model routing, and why it does not weaken this
Gaapio routes across multiple model providers — currently Anthropic, OpenAI and Google — rather than depending on one. That is a resilience and quality decision — different models are better at different parts of the work, and single-provider dependency is a risk.
The obvious concern is that more providers means more places your data goes. Which is why the zero-data-retention requirement is a condition of routing to a provider at all, not something negotiated per relationship. Every provider in the chain operates under the same commitment. The current list is published in the subprocessors section of the trust center.
For firms: independence and self-review
Public accounting firms ask a different security question, and it is not really about data. It is about independence: whether using the same tool on both sides of an engagement creates a self-review threat.
Part of the answer is architectural: a firm and its client work in separate organizations, with documents held in a per-company store and access scoped by role — so Gaapio is not acting as both preparer and reviewer inside one shared workspace. That mitigates the threat but does not by itself resolve it — the analysis also depends on what firm personnel do with the output. That is worth confirming with your own independence group, and we are happy to join that conversation — see solutions for audit firms.
Frequently asked questions
Still have questions?
Security questions are best answered by a person. Talk to us, or start with the trust center at security.gaapio.com.
Trust center: Attestations, controls, subprocessors and documentation: security.gaapio.com. This page summarizes; the trust center is authoritative.

