Share Scrub: catch secrets and personal data before you share an AI chat link

Scanning prompts locally before you send them is already crowded. Offsend, promptShield and Prompt Armour cover individuals, while Nightfall, Harmonic and Strac sell to companies. All of them check what goes into the AI. I found no product that steps in at the share button and checks the whole conversation, including the AI's replies and anything pasted earlier, before a public link is created. I also found nothing that lists your existing share links across ChatGPT, Claude and Gemini and re-scans them for secrets. The bulk-delete extensions only clean up chat history, and the built-in share-link pages are manual, one platform at a time, and don't look at content. The timing helps. ChatGPT shared chats showed up in Google in August 2025, and Claude shared chats were indexed in July 2026, with reports of API keys and medical reports in them. The Link audit is probably the strongest wedge: it is easy to explain, it is the job people are scrambling to do right after an incident, and nobody seems to sell it. There are three risks. First, the moat is thin, because any of the prompt-scanning extensions could add a share-button hook in a few weeks. Second, the platforms could add their own warnings at share time. Third, users don't trust extensions that read AI chats, after the Chrome extension caught harvesting conversations from 900,000 users. To answer the trust problem, keep the code open-source or auditable and request minimal permissions. Team patterns for small agencies, such as client names and internal hostnames, give you a defensible paid tier, because the consumer tools don't target that and enterprise DLP is too expensive and heavy for a 5 to 20 person shop. Two caveats. Each platform's share flow and link-management pages change often, so the integrations will need ongoing maintenance. And I couldn't confirm what "Muse" refers to as an AI chat platform.

• Offsend (Lab13)

  Firefox add-on. Its listing says it detects API keys, tokens, private keys, passwords and personal data locally as you type in ChatGPT, Claude, Gemini and others, and masks them with reversible mappings. It checks prompts before you send them, not shares, but it uses the same local detection engine Share Scrub would need.

• promptShield

  Free, offline Firefox add-on. It swaps names, emails, phone numbers and account numbers for stable codes before text goes to the AI, then restores the real values in the reply. It works on prompts, not share links.

• Prompt Armour

  Launched on Product Hunt. It catches and redacts sensitive data in real time inside ChatGPT, Gemini and Claude before you hit send, using regex, entropy checks for API keys and lightweight NLP, all in the browser. Its pitch to individuals is close to Share Scrub's.

• Nightfall DLP for Browsers

  Enterprise data loss prevention (DLP) extension for Chrome, Chromium and Firefox. It finds personal, payment and health data plus secrets in ChatGPT and other AI apps and redacts before you submit. Per its own documentation it only monitors Claude and does not redact there. It is sold to companies, not freelancers.

• Harmonic Security

  Enterprise browser extension and desktop agent that uses small language models to judge how risky data sent to ChatGPT, Copilot and Gemini is. It is aimed at security teams.

• Strac

  Enterprise DLP with a browser extension that warns, blocks or redacts sensitive prompts and uploads before they reach Claude and other AI tools. It covers prompts, not share links.

• AI Bulk Chat Manager and similar bulk-delete extensions

  Chrome extensions such as AI Bulk Chat Manager, AI Chat Sweep and DeclutterGPT that bulk delete or archive chats on ChatGPT, Claude, Gemini and Grok. They clean up chat history but do not list, scan or revoke public share links. They show that cross-platform extensions can manage your account on each site.

• Built-in shared-link managers in ChatGPT and Claude

  ChatGPT (Settings, Data controls, Shared links, Manage) and Claude (Settings, Privacy, Shared chats) let you list and revoke links by hand, one platform at a time, with no content scanning. This is the free baseline that the Link audit feature has to beat.

Checked 2026-10-21