Shield360 for Microsoft 365 (Office add-in)
Overview
Section titled “Overview”The Shield360 add-in brings your organisation’s AI governance into the documents and messages people write in Word, Excel, PowerPoint and Outlook. It opens as a task pane from the Shield360 button on the Home tab (Word, Excel and PowerPoint) or on the message ribbon (Outlook), and has five tabs:
| Tab | What it does | What leaves the machine |
|---|---|---|
| Check | Lists the factual claims in the selection and judges each one against your organisation’s approved sources: Supported, Unsupported or Contradicted, each with a reason. The result is labelled as an AI check. Report a wrong answer files a report for review. | The selection goes to Shield360 and on to your environment’s evaluation model. Sensitive values are masked first unless the person clears Mask sensitive values before sending. Nothing is stored. |
| Sensitive data | Scans the selection or the whole document with Shield360’s detectors, then masks or highlights what it finds. | Nothing. The scan runs in the task pane. |
| Prompts | Lists the published prompts in your Prompt Hub. The person fills in any variables, then inserts the prompt into the document or copies it into Copilot or any other assistant. | Reads prompts only. |
| Procedures | Shows your organisation’s checklist for a task. Request approval files an AI request. | The request is sent. Checklist ticks stay on the device. |
| AI tools | Shows which AI tools are approved, compared with the kinds of sensitive data found in the document. | Reads approval status only. |
The tabs
Section titled “The tabs”The person selects text (or cells) and chooses Check selection. Shield360 sends the selection to the evaluation model set up for the environment, together with the Context hub entries tagged approved-source (up to 20 sources and 60,000 characters). Each claim comes back with a verdict, a reason and the sources used. Without approved sources, claims are judged against general knowledge only, and the result says so.
Report a wrong answer asks what the right verdict was and what went wrong, and files a report for a reviewer. Only the verdicts, sources and model are sent, never the text. Copy details lets the person share the claim with a reviewer themselves.
Every check is a paid call on your evaluation model’s key, so each person may run 30 checks per 10 minutes.
Sensitive data
Section titled “Sensitive data”Scan selection or Scan document (Scan this sheet in Excel) finds API keys, private keys, passwords, connection strings, Social Security numbers, payment cards, tokens, cloud credentials, email addresses, phone numbers, IBANs and IP addresses. Mask all replaces them in place; Highlight marks them. The first 256 KB is scanned.
| Word | Excel | PowerPoint | Outlook | |
|---|---|---|---|---|
| Read the selection | Yes | Yes (cells, tab-separated) | Yes | Yes while writing; the whole body of a received message |
| Scan the document | The body | The active sheet’s used range | Every text box where the app supports it, otherwise the selection | The body |
| Mask in place | Yes, keeping formatting. Values over 255 characters or split across formatting are left for manual editing. | Text cells only, never formulas | Yes, but text is replaced per text box, which can reset formatting inside it | While writing. A received message is read-only. |
| Highlight | Yes | Yes (cell fill) | No | No |
Prompts
Section titled “Prompts”Only published prompts appear; drafts never do. Each use counts as a download from office_addin in the Prompt Hub.
Procedures
Section titled “Procedures”A procedure is a Context hub entry tagged procedure. Its name is the task, and its content is the checklist: one step per list item (- , 1. , - [ ] ) or per line. Progress is kept on the device.
Request approval asks for the AI tool, what it will be used for, why it is needed and the data involved (with the kinds found in the document), then the eight EU AI Act Art. 5 screening questions, loaded from Shield360 so they read exactly as on the web form. Every question must be answered. The result is one of:
- Filed for review, with a link to the request.
- Recorded but it cannot be approved, when an answer flagged a prohibited practice.
- Held, when the questions could not be loaded and the request went without them. A link opens the request in Shield360 with the questions showing.
See Requests from Microsoft Copilot and the Office add-in.
AI tools
Section titled “AI tools”Lists the AI tools in the AI inventory with their approval, compared with the kinds of data the Sensitive data tab found: Approved, Approved with conditions, Not approved for this data, Not yet reviewed or Not approved, with any conditions and expiry. It also shows the approved model providers when an approved-model list is enforced.
Before you deploy
Section titled “Before you deploy”Set these up in Shield360 first:
- Evaluation model (needed for Check): in the evaluation settings (
/evaluations/settings, under Developer tools → Testing), pick a provider, a model and a Vault key. See Evaluation configuration. - Approved sources (recommended): under Developer tools → Prompts & contexts → Contexts, tag the active entries Check should treat as the truth
approved-source. - Procedures: in the same place, tag active entries
procedure. - Prompts: publish prompts under Developer tools → Prompts & contexts.
- AI tools: the AI inventory and approved models already hold the decisions. When AI requests are approved with data categories, the add-in can tell which tools fit a document’s data.
Install
Section titled “Install”The add-in has two manifests: shield360-documents.xml for Word, Excel and PowerPoint, and shield360-outlook.xml for Outlook. They point at your Shield360 address, which serves the add-in under /office-addin; the manifests for Shield360 cloud point at https://app.shield360.ai. A self-hosted deployment needs manifests generated for its own address.
The add-in works on Windows, Mac and the web. It does not target Office on mobile.
Deploy to your organisation
Section titled “Deploy to your organisation”You need the Global admin or Exchange admin role in Microsoft 365. An Exchange admin also needs the Application Administrator role, or App registrations set to Yes.
In the Microsoft 365 admin center, go to Settings → Integrated apps, choose Upload custom apps, then Office Add-in, and upload shield360-documents.xml.
Just me, Entire organization, or Specific users/groups. Groups must be top-level; nested groups and security groups without mail aren’t supported.
Accept ReadWriteDocument and choose Deploy.
Upload shield360-outlook.xml the same way. It needs ReadWriteItem.
The add-in can take up to 24 hours to reach everyone. Microsoft’s requirements: a Microsoft 365 Business or Enterprise licence that includes Exchange Online with an active mailbox, and Word, Excel and PowerPoint Version 1704 or later on Windows or 15.34 or later on Mac. Tenants in sovereign or government clouds (GCC, GCC-H, DoD) deploy through the older Centralized Deployment page.
The add-in’s code updates whenever Shield360 is updated. Only a manifest change (names, buttons or permissions) needs a new upload, under Integrated apps → the add-in → Update. To remove it, select the add-in under Integrated apps and choose Remove app.
Try it yourself first
Section titled “Try it yourself first”To test before deploying, sideload a manifest:
- Word, Excel or PowerPoint on the web: Home → Add-ins → More add-ins → My add-ins → Upload My Add-in, then pick
shield360-documents.xml. - Outlook (web, Windows or Mac): open aka.ms/olksideload, then My add-ins → Custom add-ins → Add a custom add-in → Add from file, and pick
shield360-outlook.xml. It then appears in every Outlook client signed in to that mailbox.
Connect
Section titled “Connect”The first time the task pane opens, it asks the person to connect:
- Choose Sign in with Shield360. A sign-in window opens on your Shield360 address, using your company sign-in when single sign-on is set up.
- After sign-in, the add-in shows Connected as the person. It works in the Shield360 project and environment they had selected.
Signing in gives the add-in a personal token that works only for the add-in, lasts 30 days, and is revoked by Sign out. Shield360 keeps only a hash of it. Each request acts as that person with their current role, so a role change or leaving the organisation takes effect at once. If the sign-in window cannot open, allow pop-ups for Office.
Use a project API key instead is for pilots only. A key acts as the person who created it and can read the project’s Vault secrets, so never give one to many people.
Permissions
Section titled “Permissions”Each tab checks the person’s Shield360 role on every request. A tab their role does not include says Your Shield360 role does not include this. Ask an administrator.
| Feature | Permission |
|---|---|
| Check, Report a wrong answer | Run evaluations (evaluation.run) |
| Prompts | View prompts (prompt.read) |
| Procedures | View contexts (context.read) |
| Request approval | Request AI tools (intake.submit) |
| AI tools | View the AI inventory (inventory.read) |
The built-in Member and Viewer roles have everything except Run evaluations. Give people a role that has it, such as Analyst, to use Check.