HackerTrans
トップ新着トレンドコメント過去質問紹介求人

phico

no profile record

投稿

Show HN: ChatGPT gets your prompt before you hit send

chatwall.io
3 ポイント·投稿者 phico·4 か月前·7 コメント

Show HN: Restrict Kiro CLI access to your sensitive files

github.com
1 ポイント·投稿者 phico·5 か月前·0 コメント

コメント

phico
·4 か月前·議論
I don't think it's for "stealing" but more for preparing response..
phico
·4 か月前·議論
Yes indeed Grok fetches on each keystroke.
phico
·4 か月前·議論
Hey HN,

I wrote this post (and recorded the DevTools demo) to highlight a behavior many users overlook: AI chat interfaces run JavaScript that captures your keystrokes in real-time. Even if you paste sensitive data and delete it before clicking send, background telemetry or auto-save requests have often already transmitted it to the server.

To solve this, I built ChatWall. It’s a browser extension that intercepts this process using a secure overlay.

Instead of typing into the chat's native input field (where host scripts can read it), you type into our isolated extension context. It detects and replaces PII (names, emails, IBANs) with neutral tokens locally on your machine. Only the already-masked text is ever passed to the host page.

100% Local: All processing runs in your browser. We have no backend.

Open Source: The core masking logic is on GitHub so you can audit exactly what the extension is doing.

I’d love your feedback!