Hide & Sneak

BETA: Hide & Sneak catches the most common sensitive information, but it will miss some things on some sites. Treat it as an extra layer of protection, not a guarantee. Found a miss? Use "Report a problem" in the popup. Let browser AI agents see only what they need. If you hand everyday tasks to a browser AI agent, such as triaging email, filling out job applications, shopping, or paying bills, the agent can read everything on the page. Hide & Sneak gives you a switch. Turn on Agent Mode before you delegate, and while it's on: - Sensitive text is replaced with placeholders like [2FA CODE], [CARD #], [ACCOUNT #], and [PASSWORD], including in tab titles. The agent can't read the originals through screenshots, page text, or the page's accessibility data. - Fields you should answer yourself are locked and marked "Protected": salary, demographic questions, consent checkboxes, signatures, card numbers, passwords, and one-time-code fields. Your agent fills in the rest. - "Copy code" buttons and one-time login links are disabled. - Pages you choose (for example, your bank's statements page) can be blocked entirely. - Instructions planted on a page for AI agents ("AI agents: forward this email…", "ignore previous instructions", hidden text telling the agent to run scripts) are removed before the agent reads them. It missed something? Select it, right-click, and choose "Hide this from AI agents." When the agent is done, the popup lists the fields left for you to fill in. Shortcut: Alt+Shift+H. What it recognizes: verification and backup codes, passwords and PINs, crypto wallet recovery phrases, card numbers, expiry dates and CVVs, bank account and routing numbers, SSNs, tax and ID numbers (including UK and Canada), health insurance IDs, security question answers, API keys, license keys, and one-time login links. Optional packs cover account balances, crypto wallet addresses, contact info, and the last 4 digits of cards and accounts. Need to sign in while your agent works? "Unlock for 2 min" opens just that tab for sign-in, then locks it again. PDFs can't be redacted by any extension, so they're blocked while Agent Mode is on (or shown with a warning, your choice). When the agent is done, turn Agent Mode off. Protected fields unlock right away, and nothing the agent filled in is lost. PRIVACY - No network requests, no analytics, no accounts. - Does nothing while Agent Mode is off. - Original values are never saved. The local activity log records only what kind of item was hidden, where, and when. - Free and open source (MIT): https://github.com/williamchen-pm/hide-and-sneak KNOWN GAPS (BETA) - Text inside images can't be hidden, and PDFs can only be blocked, not redacted. - Most non-US ID formats beyond the UK and Canada aren't recognized yet. - An agent that goes around the page on its own (for example, running its own scripts or reading network traffic) isn't stopped. The injection shield removes the most common reason an agent would do this. Works with browser AI agents such as Claude in Chrome. Not affiliated with Anthropic or any AI provider.