Privacy Policy

PureCipher Prompt Defender helps protect users from prompt injection and unsafe browser-to-AI handoffs. It scans webpage content locally, identifies risky handoff patterns, sanitizes suspicious content in protected workflows, and provides safe review flows.

Core page and prompt processing happens locally in the browser. The current release does not include advertising, remote analytics, or default-on telemetry.

To provide its security features, the extension may process webpage content, selected or page-derived text, clipboard-related text in protected workflows, the current page hostname, and user-selected settings.

It may also create minimal intervention metadata such as event type, affected surface, severity, count-style information, hostname, optional user feedback, and local review-scoring metadata. This supports transparency in the extension popup and optional local exports.

Private drafts and selected local files are processed in the extension-owned review tab. Bundled PDF and DOCX parsers extract supported text locally; images, OCR and embedded attachments are not inspected. Drafts, extracted text and redaction choices are not uploaded or saved as durable settings. Reloading or closing clears the draft, without a secure-memory-erasure guarantee. Original files remain unchanged; downloads contain reviewed plain text only.

Custom sensitive terms persist unencrypted in local extension storage until deleted, not browser sync. Exported profiles contain those literal terms. These terms and supported credential patterns apply to local review and recognized AI text-send actions when Prompt Guard is enabled. Send-review text, masked findings and exact-text approvals stay in page memory; approvals expire after 30 seconds, one send, or a text or settings change.

Text entered in a website editor may already be visible to that website before Send. In-page review is not an isolated private composer or protection against malicious page scripts. The separate on-demand package has no automatic page, clipboard or AI-send checks.

Private drafts, pasted answers and alias mappings stay in workspace tab memory. This feature does not persist, synchronize or transmit them. Each tab has an independent session. A new draft preserves aliases; clearing, reloading, closing or reaching the selected inactivity timeout discards the session. The default timeout is 30 minutes and can be changed or disabled.

Only exact session aliases selected by the user are restored. Supported credentials remain redacted; unknown or altered markers stay unchanged. The workspace does not authenticate answer provenance or detect all personal information, and identifying context may remain.

Copying is explicit and puts the displayed result on the system clipboard, which may retain its own history. Clearing a session does not clear that clipboard. Revealed originals are hidden when the workspace is backgrounded. Nothing sends a prompt automatically.

• Sync storage: protection settings and optional-feature preferences.
• Session storage: recent intervention history and short-lived, one-time Safe Handoff or Safe Summary payloads when supported by the browser.
• Local storage: custom sensitive terms, optional structured learning signals, a local salt for hashed host indicators, and compatibility fallbacks. Optional learning signals are retained for up to 30 days unless cleared sooner.
• Memory: temporary review text and approvals, workspace drafts, answers and alias mappings, and other runtime protection information.

Version 0.2.5 does not upload raw page text, raw prompt text, clipboard contents, full browsing URLs, or page titles to PureCipher by default. It does not sell user data, share data with data brokers, or automatically transmit structured learning signals.

The release does not fetch or execute remote code or fetch remote rule feeds. Optional federated-learning features are disabled by default. If explicitly enabled and configured with an HTTPS endpoint and trusted signing key, they may send aggregate metadata to the endpoint chosen by the user. Firefox also requires applicable data-transmission permissions. Core protection works with these features disabled.

Optional aggregate payloads may include a locally generated pseudonymous client identifier, browser and extension versions, count-style intervention and feedback summaries, provider and surface categories, model and policy identifiers, sync timestamps and severity totals. They are designed to exclude raw page or prompt text, raw URLs, titles, clipboard contents, cookies and example-level event records.

If enabled, the extension can create privacy-preserving reports or structured learning batches locally. These exports are designed to omit raw page text, prompts, URLs, titles, and intervention messages, and to use hashed or per-report host indicators instead of raw hostnames.

Exports are copied locally for the user and do not leave the device unless the user chooses to share them. This is separate from explicitly configured optional aggregate delivery described above.

Pending Safe Handoff and Safe Summary payloads include received and prepared text, line offsets and removal reasons. They expire after ten minutes, are pruned during handoff operations or browser startup, and are removed when consumed. An open review page retains its displayed content until closed. Raw payloads are not added to activity history or learning exports.

The extension requests clipboard write access for user-facing actions such as copying sanitized selections and optional local reports. When clipboard protection is enabled, it may process copied or pasted text locally to remove suspicious prompt-injection lines before an AI handoff.

If a user chooses to send content to an AI provider such as ChatGPT, Claude, Gemini, Copilot, Perplexity, Poe, Grok, or another service, the resulting interaction is governed by that provider's terms and privacy practices. PureCipher can detect, pause, sanitize, or help review a handoff, but it does not control the provider.

PureCipher's use of information received from Google APIs adheres to the Chrome Web Store User Data Policy, including the Limited Use requirements.

PureCipher minimizes persistent storage of sensitive content and prefers local processing. Users can disable protection features, keep optional learning signals off, clear local learning data, and avoid sharing optional exports. No software can guarantee perfect security.

Prompt Defender is a general browser security tool and is not designed specifically for children. If a future release introduces hosted services or cross-border data transfers, this policy will be updated before those changes are released.

Material changes to extension data handling will be reflected in an updated policy before release.

For privacy or support questions, email support@purecipher.com. For security reports, email vamsi@purecipher.com. You can also visit Prompt Defender Support.