Browser extension

Overview

The Shield360 browser extension applies your organisation’s AI policy in Chrome and Edge. It works on the AI sites people use in the browser:

• AI tools. Each AI site is allowed, shown with a warning, or blocked, as your policy says.
• Prompts. A prompt that contains sensitive data triggers a warning, has the data masked, or is blocked before it is sent.
• File uploads. Files picked, dropped or pasted into a protected AI site are read on the device and warned about or blocked by the same rules.
• Company AI accounts. When the policy asks for company accounts only, a site signed in to a personal account is warned about or blocked.
• AI spend. When a budget is reached, the vendors’ API consoles are blocked or shown with a warning.
• Records. Each decision, and each prompt submitted on a protected site, is reported to Shield360 without the sensitive values themselves.

What the extension enforces is set in Govern → Policies & rules. See the enforcement policy.

Install

There are two ways to install the extension. Use device management for every browser you need to enforce on. Use a connection code for a browser someone installs by hand.

	Device management	Connection code
For	Managed Chrome and Edge on domain-joined, Entra-joined or MDM-enrolled machines, or browsers enrolled in Chrome Enterprise Core	A browser where someone installs the extension themselves
Connects with	The environment’s install key, handed to the extension by policy	A one-time code an administrator creates
Can the person remove or turn it off?	No	Yes

With device management

Force-install the extension and hand it its configuration through your device-management tool (Intune, Group Policy, Jamf or Google Admin). The pilot runbook walks through each tool.

The configuration is generated for you under Settings → Data collection → Deployment, on the Browser extension tab:

1. Create or rotate the install key at the top of the tab, so it fills in. The key is shown only once.
2. Choose Show the managed-browser policy. The panel shows the extension ID, version, whether the build is Signed, and the update URL, with ready-made configuration on tabs for Chrome — Windows, Edge — Windows, macOS, Google Admin and Linux / JSON.
3. Paste the configuration into your device-management tool.

By default the extension is installed from your Shield360 server, which hosts the signed package and the update feed that managed browsers check. When your deployment’s extension is also listed in the Chrome Web Store, the panel offers Install from: Chrome Web Store or This server. The Web Store copy needs no signed build on your server and updates itself.

Caution

Device management installs only a signed package. If Build says Not signed, the panel shows how to fix it. Chrome and Edge ignore a force-install policy set by hand on an unmanaged machine.

Managed configuration keys

The extension reads two keys from the browser’s managed storage for extensions (the “3rd party” extension policy). The generated configuration sets both.

Key	Meaning
url	The Shield360 address the browser reports to, for example https://app.shield360.ai.
installKey	A collector install key (s360i_…). The extension exchanges it once for its own device token and keeps only the token. When a token is revoked, the extension enrols again with the install key; to stop enrolments, revoke the key.

When device management sets these keys, the extension’s options page is read-only and the extension cannot be connected with a code.

Enrolment reports the browser profile’s signed-in account email as the person, so each event is attributed to someone Shield360 already knows from seats and the collector.

With a connection code

A copy installed by hand has no device management to hand it an install key, and the install key should not be given to everyone. It connects with a one-time code instead.

Create a code

Go to Settings → Data collection → Deployment, open the Browser extension tab, and under Connect a browser without device management choose Create connection code. The code looks like K7QM4-XP2RB. It is shown once, connects one browser, and lasts ten minutes. Creating a code needs the Enrol computers permission, the same as creating an install key.

Enter it in the extension

The person opens the Shield360 extension from the browser toolbar, enters the code under Connect to Shield360, and chooses Connect. The Chrome Web Store copy already knows https://app.shield360.ai; if the extension asks for a Shield360 address, enter the one the panel shows.

The extension exchanges the code for its own device token, the same token enrolling with an install key gives. Codes are case-insensitive, and spaces and dashes are ignored.

A browser connected this way keeps only the address it connected to. If its token is revoked, it cannot enrol again by itself: the popup says the browser was disconnected and asks for a new code. The person can also choose Disconnect in the popup.

Note

A browser connected with a code can remove or turn off the extension. To enforce the policy on a browser, install the extension through device management.

Your organisation’s own AI sites

AI sites your organisation adds to the policy (Your AI sites) are not known to the extension in advance. The extension asks for access to exactly those sites, and Chrome only shows that prompt after a click, so each person accepts it once per browser profile:

• The toolbar button shows !, and the popup lists the sites under Protect your organisation’s AI sites with Allow on these sites.
• Until the person allows them, a blocked site is still blocked, a site with a warning is blocked outright, and an allowed site is not checked.

No browser policy can grant this in advance. If your device management stops extensions running on internal sites (runtime_blocked_hosts), exempt the Shield360 extension for your AI sites with runtime_allowed_hosts. See Your AI sites.

What it checks

AI sites

The extension applies the AI tools policy to the AI sites in Shield360’s catalog and to your organisation’s own AI sites. A blocked site is replaced with a page saying the tool is not allowed. A site with a warning shows a page first; Continue allows the site for the rest of the browser session.

Prompts and uploads

Prompts are checked on these sites, each with its own adapter: ChatGPT, Claude, Gemini, Microsoft Copilot, Microsoft 365 Copilot Chat, Perplexity, Le Chat, DeepSeek, Grok, Meta AI and Poe. You.com, Character.AI and Google AI Studio, and your organisation’s own AI sites, are checked with a generic adapter.

The extension looks for the kinds of sensitive data in your policy: API keys, private keys, passwords, database connection strings, Social Security numbers, payment card numbers, access tokens, cloud credentials, email addresses, phone numbers, bank accounts (IBAN) and IP addresses, plus any kinds of data your organisation defines. For each kind the policy says:

Action	What happens to the prompt
Off	Nothing. A prompt with nothing in scope goes through untouched and creates no event.
Warn	The person is asked to check before sending.
Mask	Each value is replaced with a placeholder such as [EMAIL_1], then the prompt is sent. If the site will not take the new text, the prompt is blocked instead.
Block	The prompt is not sent.

Uploaded files (text, code, CSV, JSON, Word, PowerPoint and Excel) are read on the device and checked by the same rules. A file cannot be masked, so Mask stops an upload. A file that cannot be read (a PDF, an image, an archive, or one over 10 MB) gets the policy’s action for files that cannot be checked: allow (the default), warn or block. See Files uploaded to AI sites.

Company AI accounts

When the policy is set to company AI accounts only, a site whose page shows it is signed in to a personal plan is warned about or blocked. An account the page does not identify is allowed. See Company AI accounts only.

Budgets

The OpenAI platform, Anthropic console and Google AI Studio are billed per use, so the extension applies your budgets there. When the budget decision for the person is block, the consoles are replaced with an AI spend limit reached page; a warning shows AI spend is near its limit with Continue, which allows the consoles for the rest of the session. The page can offer Request an override. Chat sites, paid by the seat, are not affected.

When the policy cannot be read

Sending is allowed unless the last policy the extension saw had Block when the policy is unavailable turned on and set to block. Then sending on the protected sites is refused with an explanation until the policy can be read again.

What people see

The extension only shows something when it acts.

Situation	What the person sees
Prompt with sensitive data, Warn	This message contains sensitive data, listing the kinds found (never the values). Go back and edit, or, when the policy allows it, Send anyway with a reason, which is recorded. Otherwise: Your organisation does not allow sending this anyway.
Prompt, Block	This message was not sent, with the kinds found and Remove it from your message and try again.
Upload, Warn or Block	Check this upload (with Upload anyway when allowed) or This file was not uploaded.
Personal AI account	This is not your company account (warn) or Use your company account (block).
Pasting sensitive data	A small badge: Shield360: 1 sensitive item pasted (Email address).
Blocked AI site	This AI tool is not allowed, with If you need it for your work, ask your administrator.
AI site with a warning	Check before you continue, with Continue and Go back.
Policy unreadable, fail closed	Sending is paused.

The prompt dialogs end with Shield360 checked this on your device. Your message was not sent anywhere. The upload dialogs say the same of the file.

The toolbar popup shows the enrolment, who configured the extension (your organisation, the options page or a connection code), the signed-in user, the policy version and its main settings, the last policy sync, events waiting to send, and whether each site’s adapter is working. Sync now fetches the policy at once. The extension refreshes the policy every 15 minutes.

Privacy

• Prompt text never leaves the page. Detection runs in the page. Shield360 receives the decision, the kinds found and a keyed fingerprint of each value, never the value.
• Files never leave the device. A warned or blocked upload is reported as its type, a size range, whether it could be read, the kinds found and the decision: never its name or contents.
• Prompt and response text are included in interaction records only when your organisation keeps content (see Data retention), with sensitive values replaced by placeholders.
• A reason for sending anyway is masked before it is recorded if it contains sensitive data.

Where results appear

What	Where
Each browser	Discover → AI usage → Computers, as browser:<id>. Its page shows whether it is checking in, the enforcement it was handed, and the decisions it reported.
Warn, mask and block decisions	Detect & respond → Detections → Guard events, filtered by the source Browser extension.
Every prompt submitted on a protected site	Govern → Records → AI interactions. See AI interactions.
Installs reporting	Overview → Home, on the Fleet & setup card.

Permissions

Task	Permission
Create an install key or a connection code	Enrol computers (collector.enroll)
Set what the extension enforces	See Policies

Next steps

Enforcement policy

Set the actions for sensitive data, AI tools, uploads and accounts.

AI interactions

Search and export every prompt submitted on AI sites.

Pilot runbook

Force-install the extension alongside the collector.

Budgets

Stop paid AI use in the browser when a budget is reached.