1.2.0 — Veil now protects the well-known AI chat services out of the box: Perplexity, Grok, DeepSeek, Mistral Le Chat, Meta AI, Poe, Qwen, Kimi, Pi, DuckDuckGo AI Chat, Google AI Studio, and NotebookLM join ChatGPT, Claude.ai, Gemini, and Copilot — no per-site enable click needed for these. Your organization can still add its own custom sites, which continue to ask for your one-time approval. No other behavior changes.

The only diff vs the approved 1.0.8 is one new permission (scripting) plus the module that uses it (src/background/dynamic-scripts.ts, unminified in the source package) and an Options-page section. Why scripting is required: enterprise administrators configure, in their own self-hosted Imperum server, which internal or third-party LLM web applications this DLP extension must protect; those hostnames are not knowable at build time and differ per customer. On each configuration sync the extension calls chrome.scripting.registerContentScripts() to attach the SAME reviewed, bundled content script (content/entry.js) to those administrator-listed origins. No remote code is executed: the server returns JSON only (CSS selectors and a URL regex); the js argument is always the literal packaged file 'content/entry.js'; no executeScript, no world:MAIN on dynamic registrations (the MAIN-world transparent hook remains limited to the three hosts statically declared in content_scripts[1].matches). No silent host access: origins come from optional_host_permissions, granted explicitly per site by the user from the Options page via chrome.permissions.request(); content scripts are registered ONLY for origins where chrome.permissions.contains() is already true; match patterns are pinned client-side to https://<host>/* and a server-supplied pattern that differs is rejected rather than widened; registrations are removed immediately on chrome.permissions.onRemoved and on sign-out. Build: Node 20, npm install && node esbuild.config.mjs in packages/extension-veil (see BUILD-INSTRUCTIONS.md in the source package); the Firefox artifact replaces background.service_worker with background.scripts (same file). Source package uploaded, as for 1.0.5/1.0.8.

Single purpose: detect and redact PII before it is pasted into web-hosted LLMs (ChatGPT, Claude.ai, Gemini, Copilot). The MAIN-world content script (page-context/transparent-hook.js) is the transparent-mode hook: it only patches window.fetch, and only for the three hosts declared in content_scripts[1].matches; all other logic runs in the ISOLATED world. No remote code execution — all JS ships in the reviewed artifact; server-supplied configuration is JSON only (redaction policy + DOM selectors). Built with esbuild, deliberately unminified/readable output with source maps included. Build instructions (exact copy): Node 20; from the monorepo package: npm ci && node esbuild.config.mjs — output lands in dist/, and dist/manifest.json for the Firefox build replaces background.service_worker with background.scripts (same file). Source package available on request, as previously provided for 1.0.5. Changes vs 1.0.5: additional prompt-editor selectors for chatgpt.com's alternate composer, persistent editor re-discovery (throttled MutationObserver), per-conversation state reset, and the prompt-injection guard (regex rules; severity-gated warn/block chip; response-side canary check).

First public release on addons.mozilla.org.

Imperum Veil is a PII-redaction layer for web-hosted LLMs. It detects names, emails, IPs, hostnames, secrets, and similar identifiers in your prompt before you send it to ChatGPT, Claude.ai,Gemini, or Microsoft Copilot, and replaces them with stable placeholders so the model never sees the original values.

Requires an Imperum tenant (https://imperum.io). On first run, open the extension Options page, enter your Imperum URL, and sign in.

Compatibility: Firefox 140+ on desktop; Firefox for Android 142+.