Features
Everything for using AI at work, on one page.
Approve the tools, publish the rules, check text before it goes out, and keep a record of all of it. Runs in your browser; nothing to install.
01
AI tool register
Which AI tools your company allows, for what, and with which data.
One list of tools
Name, web address, owner, business purpose and notes for each AI tool.
Approved, restricted or under review
A clear status for every tool, with a filter and search.
Allowed data
Mark which kinds of data a tool may see: public, internal, customer, personal, financial, source code, keys.
Review dates
Set when each tool is checked again. Owners and admins are told when a review is due.
No silent overwrites
If two people edit the same tool, the second save is refused, not lost.
History per tool
Every change to a tool, who made it and when.
02
Policies and agreements
Rules people actually read, with proof of who agreed to which text.
Two starter templates
Using AI tools at work, and Personal data and AI tools. Not legal advice: edit them to fit.
Versions nobody can edit
Publishing saves a numbered version. The database refuses any change to it.
Agreement to the exact text
Each agreement stores the version, a hash of its text and the statement agreed to.
Asked again on change
A new version asks everyone again. See who has agreed to each version.
Reminders
People who have not agreed after 7 days get an alert. Reviews coming due are flagged.
Download agreements
A CSV of who agreed to what and when, for owners, admins and auditors.
03
Scanner
Check text before it goes into an AI tool. In the browser, by pattern matching.
Paste text or pick a file
Up to a million characters, or a text file up to 1 MB, read in the page and never uploaded.
21 built-in rules
13 for keys and tokens, 2 for cards and bank accounts, 6 for personal data such as Aadhaar, PAN and phone numbers. Rule set 2026.10.1.
Your confidential keywords
Up to 200 terms of your own, each with a severity: project names, clients, code words.
Masked findings
Each finding shows its rule, severity, line and a masked value, with what to do about it.
Save counts, never text
Save summary keeps the count per rule and the size. The text and the matches are never sent.
Right or wrong
Mark a finding right or wrong. Only the rule and the answer are kept, to improve the rules.
04
Risk register
Write down what could go wrong with AI, and who is fixing it.
Likelihood times impact
A score from 1 to 25, always shown with its parts, and a level read off a fixed table.
Owner, mitigation, due date
The person a risk is assigned to can update progress and add notes.
Linked to a tool
Tie a risk to a tool in the register, or keep it general.
Statuses and filters
Open, being mitigated, accepted or closed. Filter by any of them.
History
Every change and note on a risk, in order.
CSV export
Download the register. The download is recorded in Activity.
05
Alerts
The right person hears about it, in the app and by email.
Overdue risks
The owner, or the admins, when a risk passes its due date.
Reviews due
Tool reviews and policy reviews that have come due.
Policies not agreed
A reminder to each person after 7 days.
Saved scans with high findings
Owners and admins see the counts and who saved it. The text was never saved.
Choose what is emailed
Each person picks which kinds of alert also come by email.
06
Roles, activity and your data
Who can do what, proof of what happened, and the way out.
Five roles
Owner, admin, approver, member and auditor. Checked on the server for every page and action.
Two-step sign-in
Sign in with an emailed code, then a code from an authenticator app. Owners can require it for owners and admins.
Activity trail
Each record seals the one before. The Activity page checks the whole chain and shows any break.
Exports
Activity with each record's hash, saved scans as CSV, and everything as a ZIP of JSON files.
Retention
Owners choose how long saved scans and read alerts are kept, from 30 days to 2 years.
Workspace deletion
Erased 14 days after an owner asks, so there is time to cancel.
What Shield does not do.
It does not watch AI websites or apps, block or rewrite anything, or read Word, PDF or image files. It finds what its patterns describe, not meaning. The full list is on the security page.