AI & Your Data

Scramble AI is an optional, in-product assistant that lets an organizer ask in plain language (“who’s leading?”, “how many teams are registered?”) and, with your confirmation, carry out live-day actions (check a team in, move a starting hole, send a broadcast). It is turned off by default and is enabled per organization — it is off by default and can be turned off at any time.

ScrambleSync does not use your data — or anything you type to the assistant — to train, fine-tune, or improve any machine-learning model, for our own purposes or anyone else’s. This is already a binding commitment in Section 4.1 of our DPA.

The assistant is powered by Anthropic’s Claude models via Anthropic’s commercial API. Under Anthropic’s commercial terms, inputs and outputs submitted through the API are not used to train Anthropic’s models. Anthropic may retain API content briefly only for safety and abuse monitoring, not for model training.

The assistant only ever accesses your own organization’s data, and only the data needed to answer the specific question or perform the confirmed action. Under the hood it acts as an authenticated client of our own API using a short-lived, least-privilege, organization-scoped token — the same tenant isolation and permission checks that protect every other request. It cannot reach another organization’s data, and it has no standing access between requests.

We pass the model only what it needs to route your request; it fetches specific data through scoped tools rather than us shipping your whole database into a prompt.

The assistant never changes your data on its own. When it proposes an action it pauses and shows you a Confirm / Cancel preview; nothing happens until you approve it. Irreversible actions (for example, removing a team) display an explicit warning. Every action the assistant performs is written to your immutable audit log, recorded as having been done by the assistant on the confirming organizer’s behalf.

Scramble AI keeps no durable copy of your data: a pending action awaiting your confirmation expires within about five minutes, and the usage and audit records it writes contain no message content.

AI assistants can be wrong or incomplete. You are responsible for reviewing and verifying the assistant’s answers and any suggested action before you rely on or act on them. Because every mutating action requires your explicit confirmation (Section 4), the decision to proceed is always yours. ScrambleSync is not responsible or liable for any loss or harm arising from your reliance on AI-generated output that you have not reviewed, or from an action you confirm. Treat the assistant as a helpful aid, not a substitute for your own judgment as the organizer.

When Scramble AI is enabled for your organization, Anthropic, PBC (United States) acts as a sub-processor for the limited purpose of generating the assistant’s responses. It is engaged only when the assistant is in use, receives only the content of your request and the scoped data needed to fulfill it, and does not train on that content. Anthropic is listed in Section 8 of our DPA; our 30-day sub-processor change-notice applies.

Scramble AI is opt-in per organization and can be disabled at any time. Usage is rate-limited and capped by a monthly budget. If you would like it enabled or disabled, contact [email protected].