PromptGuard is a browser extension that protects you from prompt injection attacks — attempts by malicious web content to hijack your conversations with AI assistants. This policy explains what data is collected, why, and how it is handled.
When PromptGuard detects a suspicious prompt, it may collect and transmit:
| Field | Description | When |
|---|---|---|
ts | Unix timestamp of the event | Every flagged request |
url | LLM API endpoint URL (e.g. api.openai.com/…) | Every flagged request |
verdict | WARN or BLOCK | Every flagged request |
score | Numerical risk score (0–1) | Every flagged request |
matches | Pattern IDs that triggered (e.g. P001:override) | Every flagged request |
prompt | Prompt text, truncated to 2 000 chars | WARN events only (for L2 analysis) |
ua | Browser user-agent (first 120 chars) | L2 analysis requests only |
PromptGuard does NOT collect:
Flagged events are sent to the PromptGuard L2 proxy, hosted on Railway, at https://promptguard-l2-production.up.railway.app. The proxy:
Data is never sold to third parties, used for advertising, or shared with LLM providers.
chrome.storage.local and cleared when you clear extension data.chrome.storage.local): last 50 events for the popup. Clear any time via the popup's "Clear" button.sessionStorage): single-use flag set by "Send anyway". Cleared after one use or when the tab closes.PromptGuard is fully open source under the MIT License. You can inspect, audit, or self-host every component at github.com/ahmetai-cell/prompt-guard.
Privacy concerns or data deletion requests:
GitHub Issues: github.com/ahmetai-cell/prompt-guard/issues