PromptSafe catches client identifiers, financial data, and secrets in ChatGPT, Claude, Gemini, and Dola before they are sent, then keeps a record you can show as proof. Everything runs on the device.
Pasted or typed card numbers, SSNs, bank details, access keys, and more are flagged on the spot, with checksum checks to keep false alarms low.
Staff replace sensitive values with safe placeholders like [SSN_1] and keep working. Strict mode removes the "send anyway" option.
An audit log of rule names and counts, never content, an inventory of which AI tools staff use, and a one-page report for insurance renewals and client questionnaires.
Counts only. Prompt text never leaves the device and is never stored. The Team report is in development; this is an illustrative example.
AI Data Protection Summary Example Firm LLP Period 1–30 Sep Protected seats 12 / 12 Sensitive items caught before sending 47 Payment cards ......... 9 Social Security nos ... 14 Email addresses ....... 18 Access keys / secrets . 6 Redacted before sending ........ 41 Sent anyway (override) ......... 6 AI tools in use ChatGPT · Claude · Gemini · Dola Prompt content: never collected
Founding firms get early access, a founding discount, and a say in what we build next. No payment now, no commitment.
No. PromptSafe reduces accidental leaks and documents your controls. It does not guarantee protection or legal compliance, and it is not legal advice.
The extension reads text in the prompt box on supported AI sites, on your device only, to find sensitive patterns. The Free version sends nothing to us, and the Team version uploads counts only.
It uses pattern and checksum rules, so names and free-form facts in a prompt are not caught yet, and files are not scanned. Treat it as a safety net, not a substitute for a policy.