# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/rampart-browser-native-on-device-pii-radaction-50024242 # Rampart: Browser native on\-device PII radaction 62 points · 28 comments [Full discussion]() [Read original]() Category: [Safety & Privacy]() ## Skept-o-meter & Hotness Skept\-o\-meter: High\. Estimated from 3 comments\. 3 comments for the summary\. Peak rank: \#5 Time in Top 10: Not enough history Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\. 1 recorded rank observations from 2026\-10\-10T23:01:00\.95027\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\. Hotness — latest 1 recorded Hacksnap ranks: 2026\-10\-10T23:01:00\.95027\+00:00: rank \#5 Rampart claims browser\-native PII redaction at 14\.7 MB and 98\.4% recall, but the sparse discussion questions HIPAA fit and government trust, not the benchmark\. ## The brief NDS has open\-sourced Rampart, a browser\-native PII redaction system that combines deterministic regex validations for structured identifiers with a MiniLM model for names and street addresses\. It runs locally via WebGPU, ships a 14\.7 MB model, and reports 3\.9 ms median latency and 98\.4% private\-term recall across seven languages\. The project positions itself as an alpha first line of defense, not a complete compliance solution\. - Rampart pairs regex rules for SSNs, credit cards, phone numbers, routing and account numbers, emails, IPs and government IDs with MiniLM for contextual names and street addresses\. - The source reports a 14\.7 MB model, 3\.9 ms p50 browser latency on WebGPU, and 98\.4% private\-term recall across seven Latin\-script languages\. - On a 30,000\-row held\-out OpenPII test set, the source reports 98\.42% recall for Rampart versus 97\.4% for OpenAI Privacy Filter, 94\.2% for GLiNER small v2\.1, 81\.5% for a BERT\-small PII model, 65% for Microsoft Presidio and 63\.8% for AWS Bedrock Guardrails\. - The pipeline redacts text before sending and can reveal placeholders in the model reply, per the example API\. - The source labels Rampart alpha, a first line of defense, and limited to English, Spanish, French, German, Italian, Portuguese and Dutch\. ## Discussion themes Analyzed: 2026\-10\-10T23:00:42\.802283\+00:00 Analysis sample: Based on 3 of 3 usable stored comments. Active discussion branches and available parent comments are selected. This sample may omit parts of the full thread. Selected themes do not measure community opinion or how common a view is. ### HIPAA suitability questioned A commenter asks if Rampart is good enough for HIPAA; a reply says no, but supplies no reasoning or compliance detail\. Sources: [Comment 50035251]() · [Comment 50035946]() ### Government trust in data protection One commenter expresses skepticism that the current government can be trusted to protect personal information, a concern separate from Rampart's local\-processing design\. Sources: [Comment 50033952]() ## Sources & coverage AI-generated summary · 2026\-10\-10T23:00:43\.443958\+00:00 Based on 3 of 3 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened. Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.