Responsible AI
Seven principles for a tool that reads other people's documents. This concept keeps them literally: there is no model and no collection to be careless with.
Say what is connected
If no model is running, the page says so on every marked document. If one ever is, it will be named and the page will say what it is allowed to see.
Show the work
A mark without a reason is noise. Every mark carries the rule, the evidence and a next step.
A person decides
The tool points. It does not approve, reject, pay, sign or reply.
No hidden collection
This site has no text box, no upload and no account. The only form is the contact form (name, email and choices, no message box) and it stores nothing.
No invented numbers
No accuracy percentage, client count or hours saved appears here because none exists.
Plain about limits
The methodology page lists what a tool like this cannot do.
Reversible
Marks are removable and nothing is changed in the original document.
What this site does with your data
Nothing. It does not read what you click, does not send documents anywhere, does not store the choices you make in the demo, and has no model to send them to. If analytics consent is given in the site banner, only anonymous page events could be recorded, and none are sent in this concept.
Before a real model is connected
- A separate decision and a written scope of what it may read.
- Testing on real documents with a published method, and honest results.
- A named person responsible for every output that leaves the building.
- A route to correct or remove anything it has marked.