v3.0.2 — SafeAI HITL Redaction + Compliance Fixes

New: Draxion SafeAI — Human-in-the-Loop (HITL)
redaction panel. When sensitive data is detected
(SSN, credit cards, API keys, etc.), employees
now see a review panel showing the original vs
safe redacted version before sending. Employees
approve the safe version, which is submitted to
the AI tool instead of the original.

Fixed: Removed unused permissions (identity,
scripting) that caused the previous submission
to be rejected.

Fixed: Icon sizes corrected (32×32, 48×48,
128×128 px).

Fixed: Replaced innerHTML usage with safe DOM
API methods.

Fixed: Added Firefox data collection declaration
(data_collection_permissions) as required by
Mozilla policy since November 2025.

Version 3.0.0 — Complete rebuild

• New M2M authentication — connect with your
  organization's API key instead of requiring
  a dashboard login
• Real-time content analysis — detects sensitive
  data (SSN, credit cards, API keys) before it
  reaches the AI tool
• DLP block overlay — prevents submission of
  sensitive content with a clear explanation
• Prompt injection detection — catches 50+
  jailbreak and instruction-override patterns
  including encoded attacks (base64, ROT13)
• Business confidential detection — warns on
  M&A language, legal privilege, HR sensitive
  content
• Floating risk indicator — shows real-time
  risk level while typing
• Supports ChatGPT, Claude, Gemini, Cursor,
  GitHub Copilot, Perplexity, and 150+ AI tools
• Chrome MV3 architecture ported to Firefox MV2