How we keep AI safe inside your business.
AI that can read your inbox or update your CRM needs real controls, not good intentions. These are the rules every WTSync workflow follows, for every client.
- Workflows run in accounts you own, and your data is never shared with another client.
- Each workflow gets only the access it needs, read-only wherever possible.
- A named person approves anything consequential before it happens.
- Every action is logged, and you can switch any workflow off at any time.
Your accounts, your data
Each client has separate accounts, projects, connector settings and credentials. Nothing from one client's setup can be reached from another's, and our own workspace holds no client data beyond what the agreed scope needs.
Software licences and AI usage are paid by you, in accounts you own. We don't resell them or add a mark-up, and we ask for your written approval before buying any new paid tool on your behalf.
Least-privilege access
Workflows connect through dedicated service accounts with the minimum permissions they need, limited to named resources. We never run AI on someone's personal login or an administrator account.
Passwords, API keys and tokens are kept in a managed secrets store. They never go into prompts, code, chat messages or documents, and they're rotated when staff change or if we suspect exposure.
Risks that are specific to AI
The main new risk with AI is prompt injection: an email or web form that tries to give the AI instructions. Our workflows treat anything from outside your business, such as emails, forms, documents and web pages, as information to process, never as instructions to follow.
- Each workflow has a fixed list of actions it may take, and can't give itself more permissions.
- It can only send data to destinations on an approved list.
- Inputs are checked and outputs validated before they trigger anything.
- Workflows are tested with deliberately hostile examples before they go live.
A person approves what matters
A named person at your business approves, before it happens:
- any payment, or any change to payment details
- bulk messages to customers
- deleting data
- changes to who can access what
- anything else you tell us is high-impact
Workflows that message customers or touch financial information include an approval step unless you decide otherwise, in writing. Every approval is recorded.
Logs, rollback and an off switch
What each workflow does, including its inputs, outputs and approvals, is written to a log the AI itself can't change, and we review the logs monthly. Every workflow has a documented way back: an off switch, the previous version of its settings, and a way to restore records it changed. You can switch any workflow off at any time.
Tested before it touches real customers
Each workflow is tested against written acceptance criteria in a limited or test setup first. It goes live only after you've signed off, and it comes with a runbook and a short guide for your team.
Health information
We don't accept protected health information (PHI) yet. We will only take on that kind of work once business associate agreements, a written risk analysis, approved AI configurations and trained staff are all in place. Until then, health and wellness work is limited to workflows that don't involve PHI, confirmed with you in writing.
If you ever leave
Your workflows live in your accounts and stay yours. We hand over the runbooks, documentation and configurations, so nothing stops working when we do.
Questions about security? Ask us on a call, or email support@wtsync.com. Last updated 31 October 2026.