Book a pilot

Sovereignty · A practical guide

Know where your AI data goes

Every prompt carries company data: a customer’s renewal terms, a candidate’s résumé, next quarter’s numbers. This page gives you the questions to ask any AI vendor, a picture of how data should flow, and a checklist for residency, retention and logging.

Updated October 15, 2026Not legal advice

Why it matters now

Employees are already sending company data to AI tools, sanctioned or not. IBM’s 2026 Cost of a Data Breach study found that security incidents involving shadow AI more than doubled, to 43% from 20% a year earlier, among the breached organizations it studied (IBM). Gartner predicts that by 2027 more than 40% of AI-related data breaches will be caused by improper use of generative AI across borders (Gartner, Feb 2025).

Regulation adds deadlines. Most of the EU AI Act’s rules apply from August 2, 2026, and high-risk obligations for uses listed in Annex III follow on December 2, 2027 (European Commission). See the field guide timeline for U.S. state rules.

Due diligence

Twelve questions to ask any AI vendor

Ask for the answers in writing, and check that the contract says the same thing as the sales deck.

Where it runs
  1. Where are prompts, files and outputs processed and stored? Can we pin them to one region?
  2. Can we run the platform in our own cloud account for regulated work?
  3. Which model providers and subprocessors receive our data, and on what retention terms? Is zero data retention available?
What is kept
  1. Is our data used to train any model? Where does the contract say no?
  2. How long are prompts, outputs and logs kept? Can we set retention ourselves and place legal holds?
  3. What is logged for each AI action? User, model, data sources, output and approver. Can we export the logs?
Who can do what
  1. Do you support SSO and SCIM, with roles per department?
  2. Can we restrict which models each team uses, and redact personal data before it reaches a model?
  3. Which actions need a human approver? Can anyone approve their own work?
Learning and leaving
  1. If the AI improves from our feedback, is that learning private to us? Can it ever benefit another customer?
  2. Can we export everything: threads, outputs, approvals and logs, in a usable format? How is deletion confirmed when we leave?
  3. How and how fast will you tell us about a security incident involving our data?

Data flow

How data should move through a governed AI workspace

The goal is simple. Data leaves your control boundary only when policy allows, only to models you approved, and only on terms you agreed. Everything that happens is written to a record you keep.

Data flow in a governed AI workspace A person signs in through company single sign-on and runs an app in the workspace, which sits in your chosen region or your own cloud. Policy redacts personal data and checks allowed models before anything goes to an approved model provider outside the boundary, on agreed retention terms with no training. The draft returns for a named approver. Every step is saved in a record inside the boundary, with retention, legal hold and export. Your control boundary: your region or your cloud PersonSigns in with company SSO WorkspaceRuns the app; reads onlysystems it was approved for PolicyRedacts personal data;checks allowed models ApprovedmodelAgreed retention,no training Only whatpolicy allows Named approverSays yes before anything leaves The recordPrompt, context, model, output, approver.Retention, legal hold, export. Customer, candidateor system of record
Data leaves the boundary only to approved models on agreed terms. Every step is recorded inside it.

Checklist

Residency, retention and logging

Tick what is true for your current AI setup. This runs only in your browser. Nothing is saved or sent.

0 of 15 in place

Residency

Retention

Logging

How tabbr answers

Your models, your record, your learning.

tabbr routes each department to the models you approve, keeps a record of every run that you own and can export, and learns from work your teams accepted inside your company only. Your data trains no outside model, by contract. Run it in our cloud or yours.

Book a pilotRead our approach