New SafeMTS Study Highlights AI Innovations to Boost Maritime Safety
The Safe Maritime Transportation System (SafeMTS) program has released a special report detailing the results of a pilot project testing the use of large language models (LLMs) to process and analyze near-miss safety event data in the commercial maritime industry.
In March 2026, the Bureau of Transportation Statistics (BTS), the data steward for the program, hosted a stakeholder meeting to present findings from the SafeMTS Report: Applying Large Language Models to Maritime Near-Miss Safety Data Analysis.
Near-miss events are incidents in which a potential accident, injury, or hazard was narrowly avoided, and are critical for identifying early warnings and preventing serious maritime incidents. The report examines methods for automatically processing and classifying free-text near-miss descriptions to improve efficiency, accuracy, and consistency. The project tested these advanced data analysis techniques within a secure, protected environment to ensure confidentiality while enhancing the value of aggregated safety data to participating companies.
Featuring a review of anonymized participant-submitted data, the report evaluates how automated text classification, taxonomy mapping, and human-in-the-loop review can reduce time-consuming manual analysis, improve data comparability, and support more rapid identification of safety risks and operational trends across diverse datasets.
The SafeMTS team’s research included a review of existing program taxonomies and data definitions, pilot testing of secure LLM workflows, and validation by subject matter experts from both maritime operations and data science fields. In addition to findings related to classification accuracy, efficiency gains, and emerging trend detection, the report identifies current barriers to full-scale implementation, such as inconsistent data formats, incomplete records, and varying reporting cadences, and also outlines recommendations for addressing these challenges before expanding the use of LLMs within the program.
The report aims to inform maritime companies, safety managers, researchers, and regulators about the potential of advanced language-based analytics to strengthen maritime safety programs. For example, the SafeMTS team has already shared detailed safety trends that have emerged from the study as part of one-on-one meetings with commercial maritime companies currently participating in the program. By enabling faster and more consistent analysis, SafeMTS seeks to support enhanced safety management practices, more actionable industry-wide insights, and stronger prevention strategies.
Download the LLM Pilot Report from the SafeMTS website’s Publications section.
SafeMTS is sponsored by the U.S. Department of Transportation’s Maritime Administration (MARAD) and administered by the Bureau of Transportation Statistics (BTS), in partnership with the commercial maritime industry. SafeMTS provides a voluntary, confidential reporting system for near-miss safety events, enabling companies to share sensitive and proprietary data without risk of exposure. Those interested in SafeMTS can follow program updates by visiting the SafeMTS online newsroom.
As a principal federal statistical agency, BTS has the authority and obligation under the Confidential Information Protection and Statistical Efficiency Act (CIPSEA) and the agency’s authorizing statute to protect the confidentiality of submitted data, including company information, personally identifiable information, and proprietary details. Submitted identifiable data are protected from release to the public, MARAD, and other government agencies, and are exempt from Freedom of Information Act requests. To learn more, view CIPSEA and data protection at BTS.