Using the Prism console

The console at https://app.prism-dlp.com is where you choose what Prism checks, see what it found and decide what happens next. This guide goes through it page by page. For putting Prism on devices, see Set up and the rollout guides.

Changes to Settings, Destinations, Detectors and Exceptions take effect within 30 seconds for checks. Macs pick up agent settings at their next check-in, within 15 minutes.

Signing in

• Continue with Google or Continue with Microsoft (work accounts), or type your work email and choose Email me a sign-in link. The link works once and expires in an hour.
• Google and Microsoft sign-ins with the same work email join the same login.
• Someone who isn’t in an organisation yet sees “This login isn’t in a Prism organisation. Ask your Prism admin to invite you.”
• If you belong to more than one organisation (an IT partner, for example), pick one from the switcher at the top right. Each organisation gives you its own role.

Account (top right) shows how you sign in and lets you turn on two-factor sign-in with an authenticator app (1Password, Google Authenticator, Authy). We recommend it for owners and admins.

The first week

1. Leave Settings → What happens when Prism finds something on Observe. Prism records every match and interrupts no one.
2. Roll out the extension and the Mac agent to a pilot group (Set up).
3. Look through Incidents each day. Turn off detectors that are noisy for you, mark AI tools as Approved or Blocked under Destinations, and add your own detectors for project names and customer IDs.
4. After a week, switch to Warn or Enforce.

The Getting started checklist at the top of Incidents tracks these steps for owners and admins. Hide dismisses it in that browser.

Modes

If the decision model is unreachable (Settings): Allow and flag (the default) or Block. Detector matches such as keys and card numbers are acted on either way.

Incidents

Every paste, upload, typed message or file copy Prism flagged, newest first, refreshed every few seconds.

• Columns: Severity, Policy (the main reason), Channel (where it was going), User, Action (Blocked, Warned or Allowed) and Time.
• Filters:
  • Severity defaults to Flagged and detected: everything Prism would warn about or block, plus anything a detector found, even if it was allowed.
  • Test rules (not enforced) shows what test-mode detectors and test-mode blocks would have caught.
  • Personal accounts shows activity on personal AI accounts.
  • Sensor (Browser or Endpoint) and Time range (default the last 24 hours).

Click an incident to open it:

• What was pasted shows the text with every detected value removed. Prism never stores the original values. For uploads, File shows the name, type and size and whether Prism could read it (Fully, Partly or Not checked, with the reason). None of a file’s contents are kept.
• Why lists each reason. The user said shows the reason someone gave to continue.
• Detectors lists what was found and how many times. Where this content has been shows earlier moves of the same content.
• Analyst review lets analysts and above mark the decision Should allow, Should warn or Should block with a reason. It’s kept in the audit log. It doesn’t change what happened.
• Allow for a while… (admins, on blocked or warned incidents) opens Exceptions with the person, tool and data filled in.

Destinations

The AI tools Prism knows about (88 websites, 3 desktop apps and 7 command-line tools) and any you add. Each has a status:

• Watch off means Prism ignores that tool completely.
• Each entry is separate. Blocking the Claude website doesn’t block the Claude desktop app or Claude Code. To stop a whole service, set each of its entries.
• Risk notes under each tool summarise its vendor, whether it trains on chats, where data is kept and its certifications, from the vendor’s own pages.
• Add your own in any group: a website host, a Mac bundle ID or a command name.

Settings at the top of the page:

• Which AI tools people may use: Any, with content checks, or Only approved AI tools. The second refuses every tool you haven’t marked Approved.
• Suggest instead: a name and an https link. Block dialogs then say “Use {name} instead” with a button that opens it.
• Even in approved AI tools, block secrets, card and bank details and ID numbers (on by default). Turn it off only for tools you trust like your own systems.

Work accounts. Tick Work only on ChatGPT, Claude, Gemini, Copilot, Grok or Perplexity to refuse personal accounts there. Then tell Prism what a work account looks like:

• Company email domains, for Gemini and for Grok and Perplexity business accounts. Subdomains count.
• ChatGPT workspace IDs, from ChatGPT admin settings (Enterprise or Team).
• Claude organisation IDs, from Claude settings → Organization (Team or Enterprise).
• If Prism can’t tell which account is signed in: Refuse it (safer) or Allow it and record it.

Consumer Copilot always counts as personal and Microsoft 365 Copilot always as work. Prism only ever stores work, personal or unknown, never the account itself.

Browser coverage:

• AI websites above (the standard extension), or Every site, except the ones you skip. The second needs the separate Prism DLP (all sites) extension; ask us for it at hello@prism-dlp.com, and tell staff, because it checks personal browsing on managed browsers too.
• Check typed messages when they’re sent (on by default) checks what people type into AI chats when they press Enter or Send, never while they type.

Detectors

What Prism looks for. High and critical matches can be blocked in Enforce mode; low and medium matches are removed from the text and add context.

• Built-in detectors: secrets (AWS, OpenAI, Anthropic, Stripe, GitHub, Slack and Google keys, private keys, tokens, database connection strings, passwords), payment card numbers, bank details (IBAN and UK account numbers), and personal data (NHS, National Insurance, passport, driving licence, hospital and US Social Security numbers, dates of birth, email addresses and phone numbers). Turn each on or off and change its severity. Passport numbers, hospital numbers and dates of birth only match when they’re labelled, so ordinary dates aren’t flagged.
• Ignore published test card numbers (on by default) skips cards like 4242 4242 4242 4242.
• Uploaded files. Files Prism can’t read (scanned PDFs, password-protected files, .doc/.xls/.ppt, video, over 20 MB) can be allowed and recorded as not checked (the default), warned about or blocked. Keep file names (on by default) can be turned off if names might identify people; Prism then keeps only the type and size.
• Your own detectors: up to 50, each with keywords (whole words or phrases, any case) or codes (# for a digit, @ for a letter, so PRJ-#### matches PRJ-2041).
  • New detectors start in Test mode and only record what they would match (Incidents → Test rules). Turn test mode off once you’re happy.
  • Tick Block in approved AI tools too to keep a term out of every AI tool, like secrets.
• Try it: paste sample text and Run detectors to see what the settings on screen would find, saved or not. Nothing is stored.

Exceptions

Lets one person do something Prism would otherwise block or warn about, for a limited time and with a reason. Everything they do is still recorded, marked “Allowed by exception”.

1. Enter the Person (their work email) and choose an AI tool or Any AI tool.
2. Tick What it covers: Personal data, Bank details, Anything else Prism flags, or one of your own detectors. Secrets and API keys, Payment card numbers and This tool, even if it’s blocked sit under “Only if you’re sure” and are never ticked for you.
3. Choose how long (1 hour, 8 hours, 1 day, 7 days or 30 days) and give a reason. The reason is kept with every event it allows and in the audit log.

End now stops one early, within 30 seconds. Exceptions never override prompt-injection blocks, Work only rules or the fail-closed setting.

An exception only applies when the person’s email comes from your device management or their signed-in Chrome profile, never one typed into Prism’s settings, so nobody can borrow a colleague’s. That needs extension 0.1.2 or Mac agent 0.2.2 or later.

Settings

Admins can change these; everyone else sees them read-only.

• What happens when Prism finds something and If the decision model is unreachable (see Modes).
• Approved apps: your own domains, one per line (for example acme.slack.com). Data going to them is treated as staying inside your organisation.
• Desktop agent:
  • Check pastes into AI apps, and Also cover right-click, menu and drag pastes.
  • Record files copied out, to USB drives and external disks, iCloud Drive, Google Drive, OneDrive, Dropbox and Box, each with its own tick. Copies are recorded, not blocked.
  • Report which AI tools people use, on Macs and in browsers, and Show who uses each tool. Turn the second off to show totals per tool only.
• What people see:
  • The wording of the block, “not allowed” and warning dialogs. Use {data} for what was found and {destination} for where it was going; each shows a preview.
  • A Help link for “Learn more” (your AI policy page, or Prism’s page for staff by default).
  • Offer to paste with sensitive parts removed.
  • Ask for a reason before continuing, with a minimum length.
• Approved workflows: plain-language rules for movements that are fine, for example “Finance may upload invoices to netsuite.com”.

AI tools

Which AI websites, desktop apps and command-line tools are in use over the last 30 days, with active time and how many people use each. It never shows page addresses or what anyone typed. The banner counts tools in use that you haven’t approved: approve or block them under Destinations. Click a tool for its risk notes and, if Show who uses each tool is on, who uses it.

Macs and PCs lists each computer running the agent:

• Paste protection: On, or Needs Accessibility until the person allows it once.
• Agent version, with Update available when a newer Mac agent is out.
• System and Last seen. Agents check in every 15 minutes.

Set up and Integrations

Set up (admins) has the API URL, your keys and every file for Jamf, Google Admin, Windows and other MDMs. See How devices get their settings.

Integrations (admins) sends events to Slack, Microsoft Teams, Splunk, webhooks and Microsoft Sentinel, and has the Events API. See integrations.md.

People

Who can sign in to the console. Staff that Prism protects never need an account.

• Invite someone with their email and a role. The link works once, for that address, for 7 days. Resend sends a new link; the old one stops working. Invite whoever deploys Prism as an Admin so they see Set up. Only owners can invite owners or billing contacts.
• Asking to join: colleagues from your email domain who sign up themselves wait here until you Add them with a role or Decline.
• Members: change a role or Remove someone. What they did stays in the audit log. An organisation always keeps at least one owner.

Billing

Owners and people with the Billing role can subscribe and manage billing; admins can see it.

• The page shows your status (free trial, subscribed, payment failed, read-only), the price and the number of people Prism has seen this month. A person is a work email Prism saw in the month, from a check, a Mac checking in or AI tool use. Someone with a browser and a Mac counts once.
• Subscribe opens Stripe Checkout. Manage billing opens Stripe for the payment method, invoices, company details, VAT number and cancelling.
• Extend the trial by 14 days works once.
• If the trial ends without a subscription, or a payment keeps failing for 14 days, Prism stops checking and the console becomes read-only. Subscribe to carry on where you left off. Without a subscription the organisation closes 14 days later, and its data is deleted 30 days after that.

To pay by invoice, annually or for more than 250 people, email hello@prism-dlp.com.

If something isn’t working, see troubleshooting.md.

Updated 13 October 2026