Abstract
Nowadays, auto insurance companies set personalized insurance rates based on data gathered directly from their customers' cars. In this paper, we show such a personalized insurance mechanism -- widely adopted by many auto insurance companies -- is vulnerable to exploit. In particular, we demonstrate that an adversary can leverage off-the-shelf hardware to manipulate the data to the device that collects drivers' habits for insurance rate customization and obtain a fraudulent insurance discount. In response to this type of attack, we also propose a defense mechanism that escalates the protection for insurers' data collection. The main idea of this mechanism is to augment the insurer's data collection device with the ability to gather unforgeable data acquired from the physical world, and then leverage these data to identify manipulated data points. Our defense mechanism leveraged a statistical model built on unmanipulated data and is robust to manipulation methods that are not foreseen previously. We have implemented this defense mechanism as a proof-of-concept prototype and tested its effectiveness in the real world. Our evaluation shows that our defense mechanism exhibits a false positive rate of 0.032 and a false negative rate of 0.013.
Introduction
Usage-Based Insurance (UBI) has become a cornerstone of personalized auto insurance, where premiums are calculated based on real-time driving data. Insurers typically provide customers with telematics devices that plug into the vehicle's On-Board Diagnostic (OBD-2) port. These devices monitor metrics such as speed, mileage, and "hard braking" events to assess driver risk. However, this paper identifies a critical security flaw: the implicit trust these devices place in the vehicle's internal network.
The Vulnerability: Data Manipulation Attacks
The research demonstrates that an adversary with physical access to the OBD-2 port can easily manipulate the data stream. By inserting a "man-in-the-middle" (MITM) device or using an ECU simulator, a driver can feed the insurer fabricated "safe" driving data while actually engaging in risky behavior.
The authors successfully executed two types of attacks:
- Offline Replay Attack: A recorded "safe" trip is replayed to the telematics device repeatedly.
- Online MITM Attack: A custom device intercepts real-time speed data and "flattens" abrupt decelerations (hard brakes) before they are recorded.
Testing against seven commercial telematics devices from major US insurers revealed that none of the existing systems could detect these manipulations, even when presented with physically impossible data, such as sustained speeds of 157 mph for over an hour.
Proposed Defense: Physical-Cyber Cross-Verification
To address this gap, the authors propose an Enhanced Telematics Device. The core innovation is shifting the "Trust Boundary" by incorporating a tamper-resistant, 3-axis accelerometer within the device itself. This provides a source of "ground truth" from the physical world that is independent of the car's potentially compromised electronic network.
The defense mechanism uses a Mixture of Gaussian Regression Models to correlate physical acceleration with the reported vehicle speed. This statistical approach accounts for real-world noise—such as engine vibration and road conditions—that makes simple integration of acceleration data unreliable. By using a Dirichlet Process (DP) prior, the model automatically adapts to different driving environments, identifying discrepancies between reported cyber data and observed physical movement.
Evaluation and Results
The defense was evaluated using a dataset of over 90,000 records across 1,034 miles of real-world driving. The system achieved high precision in detecting fraudulent data:
- False Negative Rate (FNR): 0.013
- False Positive Rate (FPR): 0.032
Crucially, the system proved robust against unforeseen manipulation strategies because it models the characteristics of authentic driving rather than relying on known attack signatures. Even if individual data points are missed, the system consistently identifies the "events" (like hard brakes) that insurers use to determine discounts.
Conclusion
Current personalized insurance programs are highly susceptible to low-cost hardware attacks. By integrating physical sensors and robust statistical cross-verification, insurers can effectively secure their data collection process. This escalation from purely cyber-based monitoring to physical-cyber verification provides a scalable and accurate defense against insurance fraud.
Summary updated