Counting with STAR: Shipping a Privacy-Preserving Telemetry System to Millions of Users
summary
Brave builds and ships STAR, a production telemetry pipeline that uses Shamir secret sharing and oblivious PRFs to turn hashed strings into k‑threshold encrypted reports, recovers keys only when K submissions arrive, and scales to ~60 million users. Implementation is C++ in the browser with an anonymizing proxy; it proves ~24× cheaper than MPC options and unlocks opt‑in URL indexing and referral tracking. Limitations: poor fit for low‑entropy or enumeratable inputs, thresholding attacks, and a heavier verifiable STAR mode that increases latency.
AI-written summary · about
chapters (12)
- 0:00STAR and Brave intro
- 1:15Privacy-preserving telemetry use case
- 3:01Evaluating telemetry options
- 5:50Privacy goals for STAR
- 6:52How STAR’s threshold aggregation works
- 8:15STAR vs MPC advantages
- 9:18Refining STAR academically
- 10:25Deploying STAR in Brave
- 11:06Features unlocked by STAR
- 12:10STAR pitfalls and limits
- 13:57Essential telemetry additions
- 14:21Q&A
related talks
14mEndangered Privacy: Large-Scale Monitoring of Video Streaming Services
43mDARPA and Privacy Research
19mBringing Content Blocking to the Masses
27mMetadata-private data transfer
34mPrivacy Preserving Aggregation of Ad Conversions using MPC
31mTowards making private telemetry as ubiquitous as TLS
40mKeeping data private in real-time pipelines
54mAnonymity Loves Resilience: The Case of Tor