Privacy Shield for AI Chat

Anonymize sensitive text before it is sent to ChatGPT or Claude and restore it locally in the page.

Privacy Shield for AI Chat is a local layer between you and the AI models you use every day. It replaces sensitive values in the messages you type into ChatGPT and Claude with opaque aliases before the message leaves your browser, then restores the original text in the page for you. The provider only sees something like ⟦PSH_EMAIL_A1b2…⟧. You keep seeing the real email. WHAT IT PROTECTS • Emails, phone numbers, usernames • IBANs, payment cards, Swiss AHV numbers • Identity numbers and dates of birth • Names, companies, street addresses • API keys, tokens, private keys, passwords • Any custom term you add: client names, codenames, private phrases HOW IT WORKS • Runs only on chatgpt.com, chat.openai.com, and claude.ai • Rewrites outgoing chat requests inside your browser • Keeps the alias mapping on your device; wiped when Chrome closes by default • Blocks a request rather than sending it if it cannot be inspected • Shows a badge confirming each message was checked WHAT IT DOES NOT DO • No servers, accounts, analytics, telemetry, or remote code • Requests only the "storage" permission KNOWN LIMITS Detection is rule-based. Unlabeled names may be missed; add them under "Always replace". Only typed text is covered: uploads, images, voice, and tools reach the provider unchanged. Open source under the MIT license: https://github.com/Liam-G/privacy-shield-ai-chat