A recent analysis explores why companies developing autonomous vehicles (AVs) are reluctant to share safety-critical data such as accidents, near misses, and system failures, even though these datasets could help strengthen industry-wide safety, tells Tech Xplore.
The researchers interviewed employees from 12 AV firms and found two major disincentives. First, internal politics and the high cost of curating, annotating, and managing crash data make sharing difficult, even within companies. Second, the firms view this data as a competitive asset: revealing crash patterns or edge-case failures could expose their models, reduce differentiation, and weaken market advantage.
Regulatory frameworks compound the issue. For example, in the United States and Europe, AV firms must report only basic details such as date, manufacturer, or injury status, not the underlying triggering condition, sensor behavior, or system logic.
To overcome this, the authors propose practical measures. One idea is to separate proprietary elements (sensor logs, model weights) from higher-level summaries of safety events, allowing firms to disclose key insights without sacrificing intellectual property. They also suggest academic or civic intermediaries serve as neutral curators for anonymized datasets. Another possibility: regulators design standard “exam-scenarios” (e.g., pedestrian crossing under heavy rain) that all AV systems must safely navigate, and which firms must report on when deployed.
Safety-critical data beyond the few mandated fields must become more visible, while competitive logic must evolve toward shared value for public benefit. Unless that shift happens, AV safety claims remain hard to verify, and the industry risks slower progress on rare-event resilience.