Media
1 item
Description
Redacts PII in ChatGPT, Claude, and Copilot on-prem before it reaches the model; answers rehydrate in your browser. Vizco Secure Chat is a Chrome extension that sits on ChatGPT and Claude and redacts personal and other sensitive identifiers before they leave the browser. You keep using the normal ChatGPT and Claude websites. The model provider sees placeholders. You see the original names, emails, phone numbers, and other values restored in the page. This listing is for teammates and customer pilots. The extension talks only to the Vizco gateway your organization configures. It is not a consumer chatbot and it does not replace ChatGPT or Claude. WHAT IT DOES When protection is on and you send a message, Vizco intercepts the composer, sends the text to your organization's on-prem gateway, and replaces identifiers with tokens such as [PERSON_A1B2C3D4] and [EMAIL_9F8E7D6C]. That redacted text is what ChatGPT or Claude actually receives. When the model replies, the extension finds those placeholders in the page and asks the same gateway to resolve them. Originals are substituted in the DOM for display only. They are not written back into any request to the model provider. If the gateway is unreachable or returns an error, send is blocked. Nothing leaves the browser unprotected. A visible notice explains why. SUPPORTED SITES • chatgpt.com and chat.openai.com • claude.ai The ChatGPT and Claude UI stay intact. There is no side-by-side Vizco chat client. When everything is working, the main evidence the extension is running is the protection badge and the toolbar popup. WHAT YOU SEE Toolbar popup • Connection status and signed-in account • Protection on/off • Identifiers redacted in the current conversation, with original value and placeholder • Toggle between original values and “what the provider saw” • Link to the full activity log • Configured gateway URL In the page • A protection badge on ChatGPT and Claude • Rehydrated identifiers in the conversation • Hover / reveal controls so you can show the wire form the model actually received • Inline notices if protection is offline or a file could not be processed FILES Native file pickers and drag-and-drop on ChatGPT and Claude are routed through Vizco while protection is on. The file is not handed to the site until it has been processed. If file protection fails, the upload is blocked (fail-closed). CSV and other exports from the activity log contain wire-form text only — originals are not included. ACTIVITY LOG AND SIEM The activity log records redaction events on this device: timestamp, app (ChatGPT or Claude), direction, conversation, and placeholder / wire-form text. Original identifier values are not stored in that log. You can: • Filter by app • Export CSV (wire form only) • Forward the same wire-form events to a SIEM webhook (Splunk HEC, Elastic, Microsoft Sentinel, or a generic JSON/NDJSON endpoint) SIEM forwarding is fire-and-forget. A down or misconfigured SIEM never blocks chat. API keys for the webhook are kept in Chrome session storage for the browser session only, not written to disk. Administrators can pre-set the webhook URL with Chrome managed policy (siemUrl); keys are still entered locally. ACCOUNT AND VAULT Sign in through your organization's Vizco gateway (OAuth). Access tokens live in Chrome session storage and are cleared when the browser closes. Original identifier mappings live in your organization's on-prem vault (rolling retention), in one protected vault per user. The popup and the page fetch originals on demand. ChatGPT and Claude only ever receive placeholders. The gateway URL can be set on the options page, or pushed and locked by Chrome enterprise policy (managed storage key: gatewayUrl). Optional policy can also pre-provision an OAuth client id. WHO THIS IS FOR Employees who use ChatGPT or Claude in Chrome and must keep names, emails, phone numbers, and similar identifiers off the model provider's systems. Typical rollout is an unlisted or private Chrome Web Store listing, or later a force-install via Google Admin / MDM with managed configuration. REQUIREMENTS • Google Chrome (Manifest V3) • A running Vizco gateway that this extension is configured to use • Sign-in against that gateway • Use of chatgpt.com, chat.openai.com, and/or claude.ai in the browser (not the ChatGPT or Claude mobile/desktop apps) WHAT THIS EXTENSION DOES NOT DO • It does not send chat contents to Vizco as a vendor, to Google, or to any party other than your configured gateway and optional SIEM webhook • It does not download or run remote scripts; all extension code is in the package • It does not protect ChatGPT/Claude native mobile or desktop apps • It does not protect voice input • It is not a standalone chat application PRIVACY A full privacy policy is published at: https://gist.github.com/tanaybaswa/81fd26402aab1b8eb4ef2f7555c7b782 SUPPORT For install, gateway URL, and sign-in issues, contact the administrator who shared this listing. This item is unlisted and intended only for people given the link.