Protect

AI DLP: keep sensitive data out of AI tools

For the AI you allow, AILeakShield checks prompts and files before they're sent. Choose Block, Warn or Off for each type of data, and your people keep working.

Sample data shown. Select the image to enlarge it.

Why traditional DLP misses AI

Traditional DLP watches email, file shares and endpoints. A prompt is typed or pasted into a website most web filters allow. Protection has to act in the browser, before the prompt goes, and it has to work whichever account someone is signed in with.

What AI DLP does

One policy across the AI tools your people use.

One-click data rules

Block, Warn or Off for personal, health, financial, credential, classified, network, source code and legal data.

Your own sensitive terms

Add client names, project names and codenames.

Files and images

Inspect attachments, and read the text in screenshots and scanned PDFs.

Redact and send

Remove the sensitive part and send the rest, so the work carries on.

Coaching in the moment

A warning explains the risk. If you allow it, staff can continue with a written reason, and the reason is recorded.

Rules for each group

Set different rules for different teams.

Protection when offline

If the service can't be reached, rules on the device still stop common sensitive data.

Test your policy

Try a sample prompt and see exactly what would happen.

Event review

Triage what was blocked, sent anyway or flagged, with the text cleared at the end of each day.

Secure AI workspace

An optional chat workspace with approved models and the same protection.

AI agent gateway

Inspects what AI tools send to other systems through the Model Context Protocol (MCP), with the same rules.

Works where your people already work

AILeakShield protects AI sites in Chrome, Edge and Firefox through a browser extension you can deploy silently with Google Admin, Microsoft Intune, Group Policy or Jamf.

Chat assistants

ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, Grok, Meta AI, DeepSeek, Mistral and more.

Research and writing

NotebookLM, Google AI Studio, DeepL, Poe, You.com and more.

Any account

Protection applies whether someone uses a work or personal account in a managed browser.

Designed for privacy too

Security teams shouldn't have to read everyone's prompts to do this. Prompt text and written reasons are cleared at the end of each day, blocked content is never stored, and an audit trail records what happened.

Frequently asked questions

Does it slow people down?

Most prompts go straight through. People only see a message when a prompt matches a rule you've set.

Does it work with personal accounts?

Yes, in a managed browser. Protection applies on the AI site, whichever account someone signs in with.

Does it cover desktop AI apps?

Prompt inspection runs in the browser. Desktop apps and coding agents are found through your RMM or discovery scripts, so you can restrict them with your existing tools. AI agents can use the gateway.

What data can it detect?

Personal, health and financial data, credentials and secrets, classified markings, internal network addresses, source code, legal material and your own custom terms.

Is prompt text stored?

Blocked content is never stored. Other prompt text is cleared at the end of each day (UTC). Detection records, such as the category and action taken, are kept for your audit trail.

Will it catch everything?

No security tool catches everything. AILeakShield reduces the risk of sensitive data reaching AI tools, and it works alongside your training and policies.

See it on your own data

Start with a free report of the AI connected to your Microsoft 365, or book a demo.