Platforms
AI Prompt Monitor
Personal data and secrets in AI prompts and agent traffic.
All platformsEarly access
Use AI at work without leaking what matters
Prompt-Monitor checks every prompt, paste and file before it reaches ChatGPT, Copilot, Gemini or any other AI tool. Credentials, customer records and confidential documents are stopped in the browser, and your team sees what is being shared, by whom and where.
The problem
Your people already use AI every day, on dozens of tools and on sites nobody approved. Each prompt can carry an API key, a customer list, a salary file or a term sheet. Once it is sent, it sits on someone else's servers.
Blocking AI outright only pushes people to personal devices. Trusting everyone means finding out after the leak.
Prompt-Monitor lets your people keep using AI while the sensitive part stays inside.
How it works
Roll it out in an afternoon
Deploy the browser extension to Chrome, Edge, Brave and Firefox through the tools you already run: Intune, Jamf, Group Policy or the Google Admin console. A single setup file covers machines outside your MDM. Managed installs cannot be removed by the user, and you are alerted if one goes quiet.
It finds the AI tools your people use
It covers more than 140 AI tools out of the box, including assistants built into other apps such as Copilot in Outlook. It also recognises AI chats nobody has listed yet, by how the page behaves.
Every prompt is checked before it is sent
Typed text, pasted text and attached files are checked on the device in milliseconds. Depending on your policy, the person is warned and can send anyway with a reason, the send is blocked, or it is quietly recorded.
Act on what matters
Detections reach a console with severity, the person, the device, the AI tool and a masked excerpt. They can also go to Slack, Teams, email, a webhook, or a Jira or ServiceNow ticket.
Key features
Detection that understands your data
- Credentials and secrets: keys and tokens from more than 130 cloud, SaaS and AI providers, private keys, connection strings, and passwords written in plain sentences.
- Personal data with real validation: national IDs for over 40 countries, India first (Aadhaar, PAN and more), checked with the official checksums. Also card numbers, bank details, phone numbers, dates of birth, and names and addresses written in plain sentences.
- Bulk records and identity records: a pasted customer list, or a name, birth date and ID number together, is treated as the serious leak it is.
- Confidential business material: deals, unreleased results, legal matters, security incidents, strategy, HR and health records, plus confidentiality markings in 13 languages.
- Proprietary source code: code marked as proprietary, and code that uses your company's own package names.
Your own data, recognised exactly
- Confidential documents: upload a document and Prompt-Monitor recognises it when a passage is pasted, or when someone retells it in their own words but keeps its names, codes and figures. The document itself is never stored.
- Exact Data Match: your real customer or employee records are matched in any format. Values are hashed in your administrator's browser, so they never reach us.
- Confidential terms and custom detectors: project codenames, internal identifiers and your own keyword lists.
Hard to get around
Prompt-Monitor sees through the usual tricks:
- encoded text: Base64, URL encoding and escapes;
- invisible characters and look-alike letters;
- card numbers written out in words;
- leetspeak and misspellings;
- simple ciphers, such as ROT13 or text written backwards.
It also catches:
- a key split across several prompts;
- personal data that adds up over a week, a little at a time;
- password-protected archives and documents that nobody can read.
Files, not just text
Word, Excel, PowerPoint and PDF files are read on the device before upload. When you switch image reading on, screenshots and scanned PDFs are read too. A file nothing could read is reported as unread, never as clean.
Coaching, not just blocking
A warning tells the person what was found and why it matters. They can cancel, or send anyway with a reason, which goes on the record. Start in monitor mode to measure real usage, then enforce category by category.
Visibility for the security team
- A dashboard of what is leaking and from where.
- Critical and high detections in a focused feed.
- The people and tools behind the most detections.
- Coverage across every device: removed extensions, devices that stopped reporting, and machines on an old policy.
- Executive PDF reports and a full audit trail of every change.
Built for teams
- Multi-tenant workspaces with admin and read-only roles.
- Single sign-on with SAML and OIDC.
- Policies per company, team or device.
- Alert rules you build yourself, on top of the built-in ones.
Private by design
A DLP tool should not become the next place your data leaks. Prompt-Monitor enforces this:
Detection runs on the device. Prompts are checked in the browser. No AI model reads them, and nothing is sent anywhere to be classified.
You choose what is kept. Store nothing but the verdict, a redacted copy, or the full prompt, for your whole company or per team. Excerpts in the console are always masked.
Your key, not ours. Stored prompt text can be encrypted under your own AWS, Google Cloud or Azure key.
Isolation enforced by the database. Every customer's data is separated by row-level security, not just by application code.
Documents and lists stay with you. Confidential documents become fingerprints, and Exact Data Match values become hashes, before anything is stored.
Who it's for
- Security and IT teams that need AI adoption without data-loss incidents.
- Regulated industries (finance, healthcare, legal) that must show which data reaches third-party AI tools.
- Companies handling customer records at scale (BFSI, BPO, SaaS) where one pasted export is a reportable breach.
- MSSPs running AI governance across many client tenants from one console.
Other platforms
