Use AI at work. Keep your secrets out of it.
Moat checks what your people send to Claude, ChatGPT, Gemini and Copilot, and what coding agents like Claude Code and Cursor read, before it leaves the laptop. Keys, customer data and confidential files stay put, and your governance team gets the record.
Early access is open to a small group of teams.
AI in use
This week
14
Blocked
52
Redacted
3
Exceptions
- ClaudeEngineering, ProductApproved
- ChatGPTSales, Support, FinanceApproved
- Claude CodeEngineeringApproved
- DeepSeekNot approvedBlocked
Found in code
- support-assistantcalls Claude API· Ticket text, masked emails
- invoice-ocrcalls Gemini API· Invoices, no personal data
Summarise this customer list and draft a renewal email for each account.
Works with
Banning AI tools only moves the risk to personal phones. Moat lets people keep using them, and stops the few things that should never leave the company.
What slips into AI tools
Nobody means to leak anything. A key travels with an error message, a customer file goes in for a quick summary.
Keys pasted to debug an error
An API key, a database password or a token copied along with the stack trace.
Config files handed to a coding agent
A .env file or a private key read by an agent that was only asked to fix a test.
Customer data in a quick summary
Names, phone numbers, Aadhaar or PAN numbers in a spreadsheet dropped into a chat.
Private source code
Code from a client repository pasted into a personal account to get a second opinion.
Confidential documents
Contracts, salary sheets and board papers uploaded for a faster first draft.
Tools nobody approved
A new AI app adopted by one team, with no record of what it receives or where it keeps it.
How Moat works
Four parts that share one set of rules: two that sit where people work, one for the people who set the rules, and one for the people who answer for them.
Browser extension
Checks every prompt and upload before it is sent
Works in the AI assistants your people already use in the browser. When a message or a file matches a rule, Moat warns, redacts the match or blocks the send, and tells the person why in plain words.
- Text, pasted content and file uploads
- Redacts the match and lets the rest through
- Personal accounts on work devices included
ClaudeChatGPTGeminiMicrosoft CopilotMoatToday- Redacted 1 access tokenClaude · 10:42
- Stopped upload of customers_q3.xlsxChatGPT · 10:15
- Asked for a reason: Vendor_MSA_v4.pdfGemini · 09:58
- Redacted 2 phone numbersMicrosoft Copilot · 09:31
Moat checks on this device and records the rule, the tool and the time. Not what you typed.
Product preview · sample data Coding agent plugins
Stops secrets before a coding agent reads them
Coding agents read files on their own. Moat's plugins check each prompt and every file an agent is about to send. In Claude Code, Moat runs as a hook, so a blocked prompt never reaches the model.
- Claude Code, Cursor and GitHub Copilot
- Prompts, file reads and command output
- Works in the terminal and the editor
Claude Code · ~/billing-service> fix the failing invoice test ● Read(.env) └ Blocked by Moat .env holds DATABASE_URL, PAYMENTS_KEY Rule: API keys and tokens ● Read(tests/invoice.test.ts) └ Read 84 lines ● The test needs the tax rate. I will mock it instead of reading .env.Product preview · sample data Rules console
Admins decide what is allowed, per team and per tool
Start from built-in detectors, then add your own: keywords, patterns, customer names, project code names or document labels. Each rule says who it applies to, in which tools, and what happens.
- Allow, warn, redact, block or ask for a reason
- Different rules for different teams
- Approved and blocked AI tools
Rules
New rule- API keys, tokens and private keysEveryone · All AI toolsBlock
- Aadhaar, PAN and card numbersEveryone · All AI toolsRedact
- Customer list exportsSales, Support · AssistantsWarn
- Files labelled ConfidentialEveryone · All AI toolsAsk for a reason
- Unapproved AI toolsEveryone · Not on the approved listBlock
Built-in detectors
- API keys and tokens
- Private keys
- Passwords in config
- Aadhaar numbers
- PAN numbers
- GSTIN
- Card numbers
- Bank account and IFSC
- Email and phone lists
- Source code
Product preview · sample data Governance and evidence
One record of how your company uses AI
See which AI tools are in use, which code calls which models, what was stopped and which exceptions were granted. Export the record for audits and board reviews.
- AI tools and models found in your code
- Exceptions with the reason and the approver
- Exports mapped to the rules you answer to
AI in use
This week
14
Blocked
52
Redacted
3
Exceptions
- ClaudeEngineering, ProductApproved
- ChatGPTSales, Support, FinanceApproved
- Claude CodeEngineeringApproved
- DeepSeekNot approvedBlocked
Found in code
- support-assistantcalls Claude API· Ticket text, masked emails
- invoice-ocrcalls Gemini API· Invoices, no personal data
Export evidenceDPDPRBI FREE-AIISO/IEC 42001EU AI ActProduct preview · sample data
- Redacted 1 access tokenClaude · 10:42
- Stopped upload of customers_q3.xlsxChatGPT · 10:15
- Asked for a reason: Vendor_MSA_v4.pdfGemini · 09:58
- Redacted 2 phone numbersMicrosoft Copilot · 09:31
Moat checks on this device and records the rule, the tool and the time. Not what you typed.
Governance, not surveillance
People should trust Moat as much as their admins do. It checks for the risks you define and keeps as little as it can.
Checks run on the device
Rules are checked in the browser and the plugin, before anything leaves the laptop.
Events, not transcripts
By default Moat records which rule matched, the tool and the time. Not what was typed.
Content capture is opt-in
Admins can keep the matched content for chosen high-risk rules only, and people are told when that applies.
No surprises for employees
People see when a rule applies and why, and can ask for an exception instead of working around it.
Your data stays yours
Nothing is used to train AI models. Records can be kept in India.
Short retention
Records are deleted on the schedule you set, with a default you can defend.
Built for the rules you answer to
- DPDP Act and Rules
- Personal data sent to AI tools is personal data processing. The Rules' main duties apply from 13 May 2027.
- RBI FREE-AI
- Banks, NBFCs and fintechs are asked for an AI inventory, risk ratings, policies and audits.
- ISO/IEC 42001
- The AI management system standard enterprise customers increasingly ask suppliers about.
- EU AI Act
- For teams selling into Europe: transparency, AI literacy and records of how AI is used.
Moat helps you keep the evidence. It does not make a company compliant on its own, and nothing here is legal advice.
Under the hood
What Moat is being built to do. It is an early access spec, so details may change as we roll it out with our first teams.
- AI assistants
- Claude, ChatGPT, Gemini, Microsoft Copilot, Perplexity and DeepSeek in the browser
- Coding agents
- Claude Code, Cursor and GitHub Copilot, in the terminal and the editor
- Where checks run
- On the device, before anything is sent
- What it finds
- API keys and tokens, Private keys, Passwords in config, Aadhaar numbers, PAN numbers, GSTIN, Card numbers, Bank account and IFSC, Email and phone lists, Source code, plus your own keywords, patterns and document labels
- What it can do
- Allow, warn, redact, block or ask for a reason
- What it records
- Which rule matched, the tool and the time. Content only for rules where admins turn it on
- Data location
- Records can be kept in India
- Evidence
- Exports mapped to the DPDP Act, RBI FREE-AI, ISO/IEC 42001 and the EU AI Act
- Model training
- Nothing is used to train AI models
Who uses Moat
Security teams
Stop keys and customer data leaving through AI tools, without banning the tools.
Governance and compliance
Know where AI is used, set the policy once, and have the record ready when someone asks.
Engineering leads
Let engineers use coding agents with secrets and client code kept out of context.
Everyone else
Use AI at work with a clear nudge when something should not be shared, not a reprimand.
Be one of the first teams on Moat
We are rolling Moat out with a small group of teams first. Early access teams get hands-on setup from our engineers and a direct say in what we build next.
Questions about Moat
Is it legal to check what employees send to AI tools?
In India, the DPDP Act allows an employer to process employee data for employment purposes, including protecting trade secrets and confidential information, and most companies already set rules for work devices in an acceptable use policy. Moat is built to keep that checking narrow: it looks for the risks you define, records as little as possible and tells people when a rule applies. Rules differ by country, so check your own policies and local law, especially for teams outside India. This is not legal advice.
Does Moat read everything people type?
No. Rules are checked on the device, and by default Moat records only which rule matched, in which tool and when. Matched content is kept only for rules where your admins turn that on, and people are told.
Which AI tools does Moat work with?
Claude, ChatGPT, Gemini, Microsoft Copilot, Perplexity and DeepSeek in the browser, and Claude Code, Cursor and GitHub Copilot for coding. More are added as teams ask for them.
Can we block AI tools we have not approved?
Yes. Keep a list of approved tools per team. Moat can warn about or block the rest, and records which unapproved tools people tried to use.
Is Moat available now?
Moat is in early access. We are working with a small group of teams to roll it out and shape what comes next. Request access and we will set up a call.
Is Moat made by or affiliated with Anthropic, OpenAI, Google or Microsoft?
No. Moat is a Plexel product that works alongside their tools. Their names and logos belong to them.
Claude and Claude Code are trademarks of Anthropic. ChatGPT is a trademark of OpenAI. Gemini is a trademark of Google. Microsoft Copilot and GitHub Copilot are trademarks of Microsoft. Cursor, Perplexity and DeepSeek are trademarks of their respective owners. Plexel is not affiliated with, sponsored or endorsed by any of them.