The questions a bank’s security review asks, answered in order.

The webhook events your rails already emit, and the objects behind them: programs, entities, accounts, counterparties, transfers. The platform reads statuses and identifiers. It does not receive card numbers, Social Security numbers, or document images.

Your API key is stored per tenant and used only to act through your own API. Sandbox keys only until you say otherwise; a live key is refused by default. Keys are never written to logs. Operator tokens are stored as hashes.

Access is a bearer token per tenant. Every decision and every action carries the named operator who took it. Freezing an account, suspending a card account, or sending a request to a program needs a second, distinct approver.

Every disposition and every action is hash-chained to the one before it. The chain can be verified on demand and the first break, if any, is reported with its position. Any date range exports as one file for an examiner.

Drafts of the narrative are produced by Claude, and the model name and its confidence are recorded with every draft. A person signs every decision. When no model is configured the platform writes a template from the evidence, with no model and no confidence, rather than inventing either. Commercial Claude usage is billed by Anthropic and, by default, is not used to train its models. Those certifications are Anthropic's, for Claude, not Skoor's.

Skoor does not yet hold a SOC 2 report of its own. We say so plainly. What we provide instead, on request: the architecture, a data-flow diagram, and a penetration test against the deployment you will run.

We tell you what happened, when, what was affected, and what we changed, in writing, within the window your agreement sets. The record is the evidence; it is yours to read.

Anthropic (model drafts), Railway (application and database hosting), Vercel (this site), Postmark (transactional email). That is the whole list. Deployments in your own environment remove Railway from it.