About SAFER X REDACT
Staff will use AI with customer data.
Make it safer rather than forbid it.
Staff in regulated banks can paste a customer's name, account number and Ghana Card number into a cloud AI tool in seconds. A written rule is necessary, but on its own it does not change what happens at the keyboard, and most data loss prevention tools were designed long before the prompt box became a way for data to leave.
Swap, not black out
Blacking out data as [PERSON] or [ACCOUNT] often breaks the answer, and a control that gets in the way of the work tends to be switched off.
SAFER X REDACT checks what you type or paste, and the Word, Excel, PowerPoint, PDF, CSV and text files you attach. It swaps the names, account numbers and ID numbers it finds for realistic stand-ins on your own computer, before the prompt is sent to ChatGPT, Claude, Gemini or six other AI sites.
The answer stays usable, and the AI sees stand-ins in place of every value SAFER X REDACT catches.
Blacked out
Details goneDraft a dunning letter for [PERSON], account [ACCOUNT], who owes GHS 48,500.
Swapped on your computer
Shape keptDraft a dunning letter for Adwoa Boateng, account GH6015931206899, who owes GHS 48,500.
Illustrative. Each stand-in keeps the shape of the value it replaces, and the same value gets the same stand-in throughout the prompt.
Principles
Four principles behind the design.
The checking happens on your computer.
The detection engine runs on your own machine, not in the cloud. What reaches the AI site is the cleaned prompt and, for a supported file, a safe copy.
Your prompts never come to us.
The extension has no analytics, telemetry or remote code, and sends nothing to the publisher or any third party.
The store receives only what a licence needs: your name and email when you start free or buy, a one-way machine identifier for the free start, your plan and payment reference, and your activation code if you activate online. It never sees a prompt.
If the check cannot run, the prompt is not sent.
SAFER X REDACT fails closed. While protection is on, if the engine is down, the licence is not valid or the prompt box cannot be checked, text you type or paste on a supported site is blocked rather than let through unchecked. A file that cannot be checked goes only if you choose to send it.
Be plain about the limits.
Detection uses patterns, name lists and rules, with a small language model on your computer for names and places, and each catch is logged with its type and score. Some sensitive values will slip through and some harmless ones will be swapped.
Images, scanned documents, desktop AI apps and browsers other than Chrome and Edge are not covered yet. It reduces risk. It does not make anyone compliant with any law.
Who builds it
Built by a researcher who studies trust in mobile money.
Raphael Amponsah, PhD, is an AI and data product architect working on AI governance in regulated banking and on trust in mobile money. His PhD in Information Systems, from the University of Ghana Business School, examined how trust functions in Ghana's rural mobile money ecosystem and produced the Dynamic Rural Trust Model.
His published work includes academic research on trust in rural mobile money and a blog post for CGAP and FinEquity, published on FinDev Gateway: “Who really uses her mobile money account?” His publications are listed on raphaelamponsah.com and on ORCID.
“the gap between what systems claim to do and what they do in practice, and the careful work of closing it.”
SAFER X REDACT is an independent product, published by Raphael Amponsah, PhD, as a sole proprietor.
The SAFER framework
It serves one pillar of an open framework that requires no product.
SAFER is an open framework for governing AI in regulated financial institutions, which Raphael Amponsah, PhD, wrote and edits. It sets twenty controls across five pillars: Secure, Accountable, Filtered, Ethical and Resilient.
Version 1.0 was published on 2 October 2026 under the Creative Commons Attribution-NoDerivatives 4.0 licence (CC BY-ND 4.0). It is archived on Zenodo under DOI 10.5281/zenodo.23112962, which always resolves to the latest version.
Amponsah, R. (2026). SAFER: A framework for governing AI in regulated financial institutions (Version 1.0). Zenodo. https://doi.org/10.5281/zenodo.23112962
SAFER X REDACT helps an institution meet the Filtered pillar: what data reaches AI tools, and how that is enforced. It does not meet the pillar on its own. The written data rule, the regular review of what reaches AI services and the data protection impact assessment remain the institution's work. SAFER does not require SAFER X REDACT, or any other product.
Disclosure
You should know the overlap before relying on either.
SAFER's editor also sells SAFER X REDACT and carries out SAFER assessments as a consultant. Three rules keep the framework neutral:
- No control names or requires a product.
- Any tool or procedure that produces the evidence meets the control.
- Changes are published with a change log and open for comment before they take effect.
How it is paid for
Free for personal use, paid for by work licences.
The Free plan covers personal, non-commercial use on one computer, with no card needed. A new install can start with 14 days of Pro, once per person and computer. Work use needs a Plus, Pro or Enterprise licence.
Licences are annual because detectors, and the AI sites they protect, change constantly, and a redactor that is not updated decays. Licences are how SAFER X REDACT is paid for. Your prompts never reach us, so there is nothing to sell.
The privacy promise in one line
The values SAFER X REDACT swaps never leave your computer, and if the check cannot run, the prompt is not sent.
The detail is in the security and privacy notes.
Test it before you trust it.
The playground runs in your browser and sends nothing anywhere, so you can try the swaps on your own text. It does less than the app, which adds a language model for names and places.