Privacy Shield
Privacy Shield is the guardrail that protects personal information and secrets in your AI conversations. It is maintained by work.studio and is on by default.
It checks both what you send and what the AI sends back. Some items are masked so the conversation can continue; others stop the message, because sending them to an AI model is never safe.
What it does
When a message is stopped you see a short explanation instead of an answer. Nothing you typed is sent to the AI model. The value that matched is never stored in logs.
Change what it does
In my.work.studio, open AI Guardrails. For each protection you can:
- turn it off;
- choose Mask to keep the conversation going with the value hidden;
- choose Block to stop the message.
Your choices apply to your own conversations.
Your organization can enforce guardrails
An administrator can enforce guardrails for the whole organization. You can't turn an enforced guardrail off or weaken it from AI Guardrails.
How detection works
Privacy Shield uses pattern matching with validation, not guesswork:
- card numbers must pass the Luhn checksum;
- SSNs must be in a range that is actually issued;
- secrets are matched against the formats each provider uses.
This keeps false alarms low. It can still miss information written unusually, for example a card number spelled out in words. Don't rely on it as the only control for highly sensitive data.
If a check can't run, Privacy Shield lets the message through and records the error. That way an outage never blocks your work.