ORide A PrivacyPreserving yet Accountable RideHailing Service Anh

ORide: A Privacy-Preserving yet Accountable Ride-Hailing Service Anh Pham, Italo Dacosta, Guillaume Endignoux, Juan Ram´on Troncoso-Pastoriza K´evin Huguenin, Jean-Pierre Hubaux EPFL, UNIL 2017 USENIX Security Symposium Performed by Ruoxu Yang

Ride-Hailing Services (RHSs) • Millions of users, billions of rides, hundreds of cities 2

Privacy Problems For riders For drivers 3

ORide: Goals • Privacy for riders and drivers • Usability and accountability 4

Adversarial Assumptions Service Provider (SP) Honest-but-curious Active Internet Riders (R) Drivers (D) 5

Attacks Considered • Riders and drivers might assault each other • SP might perform large-scale privacy-sensitive inferences … Home/work addresses Sensitive activities • SP might perform targeted attacks Pick-up loc. and time Drop-off loc. 6

Desired Usability and Accountability Features • Riders and drivers are accountable for their behaviors - The SP can identify misbehaving users - Even in extreme cases: e. g. , kidnapping • Usability: - Easy payment - Reputation ratings for riders and drivers - Retrieval of lost items 7

Notations: ORide Overview Rider (anonymous) ACs: Anonymous credentials cert: Digital certificate SP Drivers (anonymous) cert. D ACs cert. R ACs Ride set-up Oblivious ride matching Accountability mechanisms Ride termination Payment Reputation rating 8

Oblivious Ride Matching - Intuition • Goal: Rider can select the closest driver without revealing their locations to the SP Rider (anonymous) SP Drivers (anonymous) Computation on the ciphertext locations Output of the computation Distances to drivers 9

Oblivious Ride Matching • Goal: Rider can select the closest driver without revealing their locations to the SP

11

Oblivious Ride Matching • Rider R selected driver Dwithout revealing their locations to the SP Rider (anonymous) SP Driver (anonymous) • How R and D share their locations w/ each other? • Accountability? - R and D might assault each other - R might not want to pay for the ride - Dispute on the fare 13

Payment and Reputation Rating • Goal: The two operations do not break the anonymity of the ride Rider Driver SP cert. R ACs cert. D ACs End of the day (14) Charge Rfrom his credit card (15 a) Rate D (13) Fare report: sig. R-D{day, fare, cert. R, cert. D} (15 b) Rate R 15

Protocol Analysis • Information observed by the SP Ride DB Zone of the pick-up Obfuscated pick-up time Payment DB Rider’s ID Driver’s ID Fare Day of the ride • Targeted attacks by the SP Pick-up loc. and time of aspecific rider Drop-off locations and times are never reported to the SP 16

Evaluation • Data-set: taxi rides in NYC in Oct. 2013 (15 millions rides) Census Tracts (CT) and Neighborhood Tabulation Areas (NTA) • How practical and efficient is ORide? - Per-ride overhead - Riders’ anonymity vs. drivers’ bandwidth - Effect of Euclidean distance on ride-matching optimality 17

Implementation • Feature the oblivious ride matching algorithm • SHE parameters: > 112 bits of security - Ciphertexts and keys (124 bits) … (124 bits) - Plaintexts (20 bits) … (20 bits) • In C++ using NFLlib 4096 slots • No SSE or AVX optimizations 18

Per-Ride Overhead S 1: Naïve approach S 2: Optimized approach • Measured on Intel i 5 -4200 U, 2. 6 GHz, 6 GB RAM • Bandwidth overhead • Computation time The optimized approach significantly reduces bandwidth and computation overhead 20

Riders’ Anonymity Set vs. Drivers’ Bandwidth • Zones: Queens + Bronx Brooklyn + Staten Island Manhattan CTs Manhattan NTAs Large anonymity set and reasonable bandwidth requirements 20

Conclusion • ORide: practical and privacy-preserving - Strong privacy guarantees - Negligible overhead • Still offers key RHS features: - Accountability - Easy payment - Reputation scores Source: http: //blog. cyberghostvpn. com/avoid-being-tracked-online/ 21

Thanks！