DLP for
Anthropic Claude
Claude runs in the browser and as a native desktop app. Without DLP controls, confidential data flows freely into both. Aona detects, redacts, and hard-blocks sensitive data in real time, across Claude on the web and the Claude desktop app that network and CASB tools never see.
What Gets Leaked into Claude
These four data types appear most frequently in Claude prompts and uploads, and create the most significant compliance and competitive risk.
Customer PII
Names, email addresses, phone numbers, account details, pasted into Claude during customer service, sales, or data analysis tasks.
Financial Records
Revenue data, forecasts, earnings reports, payment card numbers, and budget documents shared with Claude before public disclosure.
Source Code
Proprietary algorithms, API keys, database credentials, and internal codebases submitted to Claude without restriction.
Internal Strategy Docs
Board minutes, M&A information, competitive strategy, and legal correspondence dropped into Claude for drafting or summarisation.
How Aona DLP Works
Three-stage pipeline from prompt to protection, all in real time, before data leaves your environment.
Real-Time Prompt Inspection
Aona intercepts every prompt and file upload before it is submitted to Claude, in the browser and in the native desktop app. The content is inspected in milliseconds, with no latency impact on the employee experience.
Classification
Aona classifies the prompt content server-side against your data policies, identifying PII, financial data, source code, and custom data types using AI-native pattern recognition.
Block or Redact
Depending on your policy, Aona hard-blocks the prompt or automatically redacts the sensitive portion, in real time. There is no acknowledge-and-continue override.
Where Aona Covers Claude
Aona DLP works across every surface Claude runs on, including the native desktop app that network and CASB tools cannot inspect.
Claude (web)
Anthropic
Claude desktop app
Anthropic
ChatGPT
OpenAI
Microsoft Copilot
Microsoft
Google Gemini
Perplexity
Search AI
Mistral
Mistral AI
+ 5,600 more
All AI tools
How Aona DLP for Claude Works
Real-time prompt and file inspection across the browser and the native desktop app, protecting data before it ever leaves your organisation.
Deploy on Every Surface
The Aona browser plugin (Chrome, Edge, Firefox) covers Claude on the web, while the Windows and macOS native endpoint app intercepts the Claude desktop app on the device. One policy, both surfaces, including the desktop app network and CASB tools miss.
Inspect Every Prompt and Upload
Aona intercepts prompts and file uploads before they reach Claude. Every prompt is scanned server-side for PII, financial data, source code, healthcare records, and custom data types, in real time, with zero latency impact.
Hard-Block or Redact
Depending on your policy, Aona hard-blocks sensitive prompts with no acknowledge-and-continue override, or redacts the sensitive portion of an uploaded DOCX or Excel file while preserving layout. Full audit log captured for every event.
of employees use unsanctioned AI tools
Gartner, 2025
average cost of an AI data breach
IBM Cost of a Data Breach Report, 2024
Network and CASB tools miss the Claude desktop app
Many engineering and analyst teams run the Claude desktop app rather than the web version. It runs natively on the device, so prompts typed into it never pass through a proxy that network DLP or CASB tools can read. Aona's Windows and macOS native endpoint app inspects the desktop app directly, so the same hard-block and redaction policy applies whether an employee uses Claude on the web or the desktop app, with no blind spot.
Claude Projects and MCP connectors pull company data into the model
Claude's most useful features for teams are also the ones that move the most sensitive data. Projects and MCP connectors both load real company content into a conversation.
Projects and knowledge bases
Claude Projects let employees attach files and reference material so Claude has standing context for a body of work. Teams routinely load contracts, customer records, financial models, and internal documents into a project knowledge base. Aona inspects the content going into a project the same way it inspects a prompt, so sensitive data is caught before it is added.
MCP connectors
Through the Model Context Protocol, Claude can connect to internal systems and pull live company data into a conversation on demand. That is powerful, and it means an employee can surface regulated material in a single prompt. Aona inspects what an employee submits to Claude in real time, so the policy applies even as connectors widen what Claude can reach.
Claude runs in the browser and as a native desktop app
Employees use Claude two ways, and only one of them is visible to network DLP and CASB tools. Aona covers both.
Claude in the browser
Most people first reach Claude on the web, pasting documents and asking it to draft, analyse, and summarise. Aona's browser plugin for Chrome, Edge, and Firefox inspects every prompt and file upload before it reaches Claude, blocking or redacting sensitive data with no acknowledge-and-continue override.
The Claude desktop app
Engineers and analysts often prefer the Claude desktop app, which runs natively on Windows and macOS. Prompts typed into it never pass through a proxy that network DLP or CASB tools can read. Aona's native endpoint app intercepts the Claude desktop app directly on the device, so the same policy applies with no blind spot.
DLP for other AI assistants
Aona applies the same real-time, server-side DLP across every major assistant your team uses. Explore the tool-specific guides.
DLP for ChatGPT
Cover ChatGPT Enterprise, Team, and free, plus custom GPTs and memory, with one consistent policy.
DLP for Microsoft Copilot
Cover Microsoft 365 Copilot, Copilot in the browser, and the Copilot desktop app intercepted by Aona's native endpoint.
DLP for Google Gemini
Cover Gemini on the web, Gemini across Google Workspace, and Google AI search with one consistent policy.
Turn DLP policy into a real AI governance program
Use these templates and resources to define what data can reach AI tools, document controls, and create the audit trail security reviewers expect.
AI Security Audit Checklist
Audit AI tool usage, data flows, prompt controls, and governance gaps before sensitive data leaves your organisation.
AI Incident Response Playbook
Give security teams a repeatable process for investigating AI data leakage, prompt abuse, and policy violations.
AI Security Statistics 2026
Use current benchmarks on AI risk, leakage, and adoption to build urgency with leadership and procurement teams.
Frequently Asked Questions
Stop Claude Data Leaks Today
Start your free 90-day trial. Full DLP coverage for Claude on the web and the Claude desktop app, plus 5,600+ AI tools in the catalog.