My Precious Crash Data: Barriers and Opportunities in Encouraging Autonomous Driving Companies to Share Safety-Critical Data

Hauke Sandhaus, Angel Hsing‐Chi Hwang, Wendy Ju, Qian Yang

Abstract

Safety-critical data, such as crash and near-crash records, are crucial to improving autonomous vehicle (AV) design and development. Sharing such data across AV companies, academic researchers, regulators, and the public can help make all AVs safer. However, AV companies rarely share safety-critical data externally. This paper aims to pinpoint why AV companies are reluctant to share safety-critical data, with an eye on how these barriers can inform new approaches to promote sharing. We interviewed twelve AV company employees who actively work with such data in their day-to-day work. Findings suggest two key, previously unknown barriers to data sharing: (1) Datasets inherently embed salient knowledge that is key to improving AV safety and are resource-intensive. Therefore, data sharing, even within a company, is fraught with politics. (2) Interviewees believed AV safety knowledge is private knowledge that brings competitive edges to their companies, rather than public knowledge for social good. We discuss the implications of these findings for incentivizing and enabling safety-critical AV data sharing, specifically, implications for new approaches to (1) debating and stratifying public and private AV safety knowledge, (2) innovating data tools and data sharing pipelines that enable easier sharing of public AV safety data and knowledge; (3) offsetting costs of curating safety-critical data and incentivizing data sharing.

Introduction

While sharing autonomous vehicle (AV) crash data could save lives, companies often treat this data as a "competitive edge" rather than a public good. In the world of autonomous driving, "corner cases"—rare, dangerous scenarios like a pedestrian lying in the road or a vehicle overtaking in a tunnel—are the hardest to solve. Training models on these events is essential for safety, yet this data is currently siloed within individual companies.

While historical safety innovations like the three-point seatbelt were shared freely to save lives, modern AV companies are reluctant to share their "precious" crash data. Previous research focused on building better sharing tools (encryption, blockchains), but the "data drought" persists. This study investigates the human and organizational reasons why industry insiders are holding back, even when they acknowledge that sharing would benefit the industry as a whole.

Methods

The researchers conducted a qualitative study, performing semi-structured interviews with 12 industry insiders from 12 different organizations. These participants represented a broad spectrum of the industry, including research engineers, data scientists, and infrastructure managers from both "Classic Automotive" (e.g., VW, Mercedes) and "Specialized AD Tech" firms (e.g., Waymo, Cruise).

Table 1: Overview of the 12 industry participants interviewed for the study.

The team used inductive coding to identify themes and journey mapping to trace how data moves through an industrial pipeline. This allowed them to see exactly where "knowledge leakage" occurs during the data curation process.

The Data Landscape

The paper analyzes the landscape of existing datasets and how they are utilized by professionals. This includes public/academic datasets like CODA (real-world corner cases) and StreetHazard (simulated anomalies), as well as regulatory data such as California DMV incident reports.

Crucially, the authors categorize internal industry data sourcing into three primary buckets: Proactive Identification, Reactive Discovery, and Data Enrichment.

Table 2: Categorization of internal industry data sourcing strategies.

Key Findings: Data as a Mirror of Design

The study reveals that the real barrier to sharing isn't technical or privacy-related; it's that safety-critical data inherently leaks a company’s proprietary "secret sauce."

The "Secret Sauce" Leak

Safety data isn't just "raw" information. How a company labels a crash reveals their internal "safety paradigm." For example, if a company specifically tracks "persons lying on the ground," it reveals a specific failure mode they are worried about and how they’ve designed their detection triggers. Sharing data risks revealing a company's ML architecture, sensor placement, and even the weaknesses of unreleased prototypes.

Privacy vs. Competition

While companies often cite privacy (GDPR) as a reason not to share, interviewees admitted this is a "surmountable" barrier. Technologies like face-blurring and homomorphic encryption exist; the real hesitation is purely competitive. The study identified four specific types of "Design Knowledge" that companies fear losing:

  1. Safety Paradigms: How they define and operationalize "safety" at scale.
  2. ML Infrastructure: Insights into sensor design and data structures.
  3. Failure Modes: A roadmap of exactly where their current system is most vulnerable.
  4. Liability Issues: Data that could be used against them in legal battles.

Discussion and Future Work

There is a sharp divide in the industry: while practitioners recognize the "economic absurdity" of 100 companies driving the same streets to collect the same data, the "secrecy mentality" currently outweighs collective efficiency.

Future work should focus on:

  • Standardized Simulation: Building shared virtual platforms where companies can test safety without sharing raw data.
  • Academic Intermediaries: Using universities as "trusted third parties" to aggregate data and release anonymized, high-level insights.
  • Policy Intervention: Informing mandates (like the EU Data Act) to distinguish between "safety-essential" knowledge (which should be public) and "competitive design" insights (which can remain private).

The authors acknowledge a Western bias in the current sample and suggest that future research should explore different cultural or regulatory attitudes in other major AV markets.

Summary updated

Abstract

Safety-critical data, such as crash and near-crash records, are crucial to improving autonomous vehicle (AV) design and development. Sharing such data across AV companies, academic researchers, regulators, and the public can help make all AVs safer. However, AV companies rarely share safety-critical data externally. This paper aims to pinpoint why AV companies are reluctant to share safety-critical data, with an eye on how these barriers can inform new approaches to promote sharing. We interviewed twelve AV company employees who actively work with such data in their day-to-day work. Findings suggest two key, previously unknown barriers to data sharing: (1) Datasets inherently embed salient knowledge that is key to improving AV safety and are resource-intensive. Therefore, data sharing, even within a company, is fraught with politics. (2) Interviewees believed AV safety knowledge is private knowledge that brings competitive edges to their companies, rather than public knowledge for social good. We discuss the implications of these findings for incentivizing and enabling safety-critical AV data sharing, specifically, implications for new approaches to (1) debating and stratifying public and private AV safety knowledge, (2) innovating data tools and data sharing pipelines that enable easier sharing of public AV safety data and knowledge ; (3) offsetting costs of curating safety-critical data and incentivizing data sharing.
BibTeX
@inproceedings{sandhaus2025my,
  author = {Hauke Sandhaus and Angel Hsing‐Chi Hwang and Wendy Ju and Qian Yang},
  title = {{My Precious Crash Data: Barriers and Opportunities in Encouraging Autonomous Driving Companies to Share Safety-Critical Data}},
  booktitle = {Proceedings of the ACM on Human-Computer Interaction},
  year = {2025},
  doi = {10.1145/3757493},
}
Figure