Bug fixes to auto-redaction system.

Backend Fixes
Safari support

Changes since last submission (0.46.47 → 0.46.55):
- Ambient AI knowledge base (137 services profiled, capture enrichment, popup risk display)
- Agentic AI category (16 standalone + 10 capability-flagged)
- Path-based service matching (Perplexity Computer, Bedrock, Copilot, Zapier agents)
- Perplexity Computer detector skip
- Remote config foundation (heartbeat delivery, RemoteConfigManager)
- Config override policy (enforced/default/reset) — feedback button wired as proof
- Credit card risk score fix (38→40) and modal risk level alignment
- 146 hostnames across 16 categories (was 132/15)

Bug fixes and expanded coverage

Bug Fixes:
Overly sensitive NER
Intervention modal visual glitch

Fixed issues with overly aggressive NER false positives impacting user experience

Bug fixes to all primary AI Chat service sentinel.
Improved intervention and editing modal experience
Advance NER PII detection enhancements, all in browser. No data leaves the device for assessment.

Vireo Sentinel Extension Release Notes

Versions 0.43.35 - 0.43.40

Version 0.43.40 (March 5, 2026)

Firefox Magic Link Activation - Fixed ✅
- Firefox users can now use one-click email activation links (previously failed)
- Manual configuration simplified - just paste the single license key from your email
- Improved activation reliability with automatic extension detection



Version 0.43.39 (March 5, 2026)

New User Onboarding - Enhanced 🎉
- Added prominent activation banner in extension options page
- Clear step-by-step instructions for license activation
- Guides users to use magic link from email (easiest method)
- Banner auto-hides after successful configuration



Version 0.43.38 (February 24, 2026)

Bug Reporter Icon - Fixed 🔧
- Fixed broken Vireo logo on floating feedback button
- Now displays correctly on ChatGPT, Claude, Perplexity, and Gemini



Version 0.43.37 (February 24, 2026)

Shadow AI Discovery - Major Expansion 🚀
- Service tracking expanded from 42 to 101 AI tools across 14 categories
- Added 5 new categories: Marketing AI, Legal AI, HR & Recruitment, Customer Support, Financial AI
- Enhanced coverage in existing categories (Coding, Writing, Research, Design, Sales, etc.)
- New services include: Replit, Sourcegraph, CodeRabbit, QuillBot, Harvey AI, Beamery, Ada, Ramp,
and many more
- Automated manifest generation system for easier maintenance

Tracked Categories (14 total):
- Search, Coding, Writing, Research, Design, Sales & CRM
- Productivity, Communication, Analytics, Marketing
- Legal, HR & Recruitment, Customer Support, Financial



Version 0.43.36 (February 24, 2026)

Branding Update 🏷️
- Rebranded "Secondary Services" to "Shadow AI Discovery"
- Better market positioning and clearer value proposition
- Updated UI labels and messaging throughout dashboard



Version 0.43.35 (February 18, 2026)

Intervention Editor - UX Improvements ✨

File Handling Fixed:
- File drops into Claude editor now correctly show attachment chips
- Attached files section displays all pending file attachments with names and sizes

Editor Clarity:
- Added context-aware subtitles explaining what happens with your content
- Blue banner clarifies that editor replaces AI chat input until you click Submit
- "Your prompt:" label makes purpose clearer

Stability Improvements:
- Fixed input hang after dismissing editor without submitting (~1-in-5 occurrence)
- Prevents duplicate intervention when pasting into editor modal
- Added cancel confirmation with optional clipboard save

New Features:
- Copy to clipboard button always available in editor
- Cancel confirmation offers to save prompt to clipboard before discarding

Version 0.43.35 — Release Notes

This release improves the intervention editor experience across all four supported AI platforms
(ChatGPT, Claude, Perplexity, Gemini).

Intervention editor redesign (v0.43.34)

• The editor subtitle is now context-aware, clearly explaining whether content was intercepted
  from a paste or from typing
• A blue instructional banner makes it explicit that nothing is sent until the user clicks Submit
• An "Attached files" section now displays pill chips for any files pending attachment, keeping
  users informed of what will be sent
• A persistent "Copy to clipboard" button is available throughout the editor session
• When closing the editor without submitting, a confirmation dialog offers the option to copy the
  prompt to the clipboard before discarding, preventing accidental data loss
• Cancelling or closing the editor no longer leaves the AI input in an unresponsive state (root
  cause: blocking flags were not reset on dismiss)
• The paste interceptor no longer fires a second intervention overlay when pasting inside the
  editor itself

File drop fix for Claude editor (v0.43.35)

• Dropping a file into the Claude intervention editor now correctly displays the file chip in the
  "Attached files" section
• Fixed a capture-phase event listener that was intercepting all drop events before the editor
  modal's own handler could process them

New permission: clipboardWrite

• Formally declared in the manifest to cover the copy-to-clipboard features added in this release.
  Used only on explicit user action (clicking Copy or confirming clipboard save on cancel). No
  background clipboard access.

Changes Since v0.43.30

v0.43.31 - Risk Assessment Fix

Fixed: Risk recalculation after user interventions
- Previously: Risk levels remained unchanged after users edited content or accepted auto-redaction
- Database showed 0 risk reductions despite 862 interventions
- Solution: Added automatic risk re-assessment on modified content after edit/redact actions
- Result: Accurate measurement of intervention effectiveness and risk reduction

v0.43.32 - Enhanced Floating Feedback Button

Added: Draggable positioning & hide/show toggle
1. Drag-to-Reposition:
- Click and drag button anywhere on screen
- Position persists across page reloads and sessions
- Constrained to viewport bounds
- Visual feedback during drag
2. Hide/Show Toggle:
- New Settings section in extension popup
- Toggle applies instantly to all AI platform tabs
- Setting persists across sessions
- Default: visible

v0.43.33 - Smart Boundary Detection

Enhanced: Intelligent expansion direction for floating button
- Problem: Button expanding near right edge would go off-screen
- Solution: Automatically detects available space and expands in optimal direction
- Near right edge: Expands leftward
- Plenty of space: Expands rightward (default)
- Added: Viewport boundary detection
- Button automatically repositions when viewport changes (e.g., dev tools opening)
- Stays visible within viewport boundaries
- Removed noisy debug logging



Summary: Three bug fixes and UX enhancements focused on accurate risk measurement and improved floating feedback button usability with
intelligent positioning and expansion behavior.

Release Notes: v0.42.1 → v0.43.30

ChatGPT Improvements

• Fixed file attachment from Content Editor Modal - Files dropped/pasted into the modal now properly attach to submissions (v0.43.23)
• Fixed submission blocking - Resolved race condition where extension would block its own approved submissions after file attachment (v0.43.24)
• Fixed async file upload handling - Extension now properly waits for large file uploads to complete before allowing submission, using intelligent input
  monitoring instead of arbitrary timeouts (v0.43.25)
• Fixed double acknowledgment popup - File acknowledgment modal now appears only once when using Content Editor Modal (v0.43.26-27)
• Fixed override bypass - Content Editor Modal no longer re-appears after providing override approval (v0.43.21)
• Fixed risk indicator cleanup - Risk badges/outlines now properly clear after submission (v0.43.20)

Gemini Improvements

• Added drag & drop education - Users attempting to drag files now see clear instructions explaining Gemini's platform limitation and the correct workflow
  using the ⊕/📎 button (v0.43.28-29)
• Fixed drag UI state - Gemini's "Drop files here" visual feedback now properly clears after showing the informational modal (v0.43.30)
• Separated workflows - Drag & drop shows only informational modal, button-based uploads show only acknowledgment popup (v0.43.29)
• Fixed risk indicator cleanup - Risk badges/outlines now properly clear after submission (v0.43.20)

Perplexity Improvements

• Fixed line break preservation - Multi-line content with Shift+Enter now preserves formatting when pasting additional content, handling Perplexity's Lexical
  editor structure (v0.43.19)
• Fixed risk indicator cleanup - Risk badges/outlines now properly clear after submission (v0.43.20)

Claude Improvements

• Removed warning banner - Cleaner interface without intrusive top banner (v0.43.18)
• Slimmer Content Editor - Removed instructions panel for more focused editing experience (v0.43.22)

Cross-Platform UX Improvements (All Detectors)

• Auto-expanding textarea - Content Editor Modal textarea now expands from 200px to 500px based on content for better editing experience (v0.43.22)
• Redaction preview highlighting - Sensitive data in "Original" column now highlighted in red for easier matching with redacted versions (v0.43.22)
• Consistent risk indicator cleanup - All platforms now properly clean up visual indicators after submission (v0.43.18-20)

Key Technical Improvements

• Per-Prompt Acknowledgment Pattern - File acknowledgment required once per prompt, subsequent files in same prompt don't re-prompt
• Input Monitoring System - Intelligent detection of submission completion or new prompt start for proper flag lifecycle management
• Platform-Specific Handling - Tailored workflows for each platform's capabilities (e.g., Gemini drag & drop limitation)

Summary of changes since v0.40.24:

1. +26 AI platforms (Grammarly, Canva, HubSpot, Notion AI, Loom, etc.)
2. Conversation lifecycle tracking across browser sessions
3. Enhanced PII detection with deterministic validation
4. Live risk assessment in content review modal
5. Critical fixes: License errors, dark mode, API connectivity
6. Security updates: All dependencies updated to latest stable