PureCipher Prompt Defender

Intercepts browser-to-LLM handoffs, sanitizes prompt-injection content, and guards AI prompt submission flows.

PureCipher Prompt Defender adds a browser-side trust boundary when webpage content moves toward supported AI services. Core protections: - detects hidden or low-visibility prompt-injection patterns in webpage content - sanitizes suspicious text before copy, paste, handoff, or prompt submission - pauses risky page-to-service controls after hidden prompt-like content is found - provides Open Safe Handoff and Open Safe Summary review flows - explains removed lines in the review and separates earlier page cleanup - supports a preferred AI destination and a configurable selection-review shortcut; you choose when to paste and send - pauses recognized text-send actions for supported credentials and saved confidential terms, with masked findings and explicit redaction choices - applying review choices updates the draft without sending; edits invalidate the one-send approval - offers local private drafting and supported document-text extraction; original files are not sanitized - shows recent interventions without keeping raw page text in long-lived history Current product boundary: - works on regular http:// and https:// webpages - protects webpage controls and prompt submission on supported AI websites - supports browser AI APIs that are exposed to webpage JavaScript - does not claim control over browser-owned native surfaces outside webpage content - text entered into a website editor may already be visible to that website; attachments and background requests are not covered by the text-send check Privacy: - core page and prompt processing occurs locally in the browser - raw page text, raw prompt text, and clipboard contents are not uploaded to PureCipher by default - privacy-preserving reports and structured learning batches are local, opt-in, and user-initiated - structured learning signals are off by default Keep the context. Choose what you share. Prepare prompts in a private workspace with consistent aliases for clients, people, projects and other selected values. Bring the AI answer back and choose which originals to restore locally. Review the exact result before copying; nothing sends automatically. - Keep the same aliases across multiple drafts in one session. - Choose categories and select confidential text yourself. - Restore exact session aliases selectively; supported credentials stay redacted. - Clear the session yourself or choose an inactivity timeout. Drafts and mappings stay in the workspace tab. Closing or reloading loses them. Detection covers selected literal values and supported credential patterns, not all personal data. Context can still identify someone. Use answers from the same session; unknown or altered markers are not guessed. Copying uses the system clipboard.