SAFE CROSS Act

• Authority: United States Congress
• Parent authority: Federal government
• Jurisdiction: United States
• Collections: U.S. federal laws
• Most recent activity: Proposed
• Most recent activity date: 2025-05-30

Risk subdomain coverage

All other risk subdomains had no coverage in this document.

Risk Coverage Summary

This document does not substantially address any specific AI risk domains from the MIT taxonomy. It is a procedural bill mandating a study on AI-enabled sensors for rail crossing safety, rather than a document that governs AI risks directly. The study itself may eventually examine safety and robustness issues, but the bill text does not engage with specific risk categories.

Sector coverage

Good coverage — Comprehensive governance measures or mitigations with clear procedures relating to the sector are explicitly described.

• Trade, Transportation & Utilities

Minimal coverage — Brief mention but no specific mitigations, recommendations or governance measures relating to the sector are described.

• Public Administration (excl. National Security)

All other sectors had no coverage in this document.

Sectors Summary

The document primarily governs AI use in the Transportation subsector, specifically addressing rail crossing safety technology. It directs the Federal Railroad Administration to study AI-enabled sensors for reducing pedestrian and traffic accidents at rail crossings, with results intended for use by governmental entities regulating rail crossing safety and private entities maintaining such safety technology.

Legislative Status

Hard Law

Legislative Status Summary

This is a Congressional bill that, if enacted, would create legally binding obligations for the Federal Railroad Administration to conduct a study and publish results within specified timeframes.

Actors

Proposer

• Entity Types: AI_Governance_Actor
• Entities: United States Congress; Senate and House of Representatives
• Summary: The bill is proposed by Congress as indicated by the enactment clause and the legislative format of the document.

Target

• Entity Types: AI_Governance_Actor
• Entities: Federal Railroad Administration; Administrator of the Federal Railroad Administration; Associate Administrator for Railroad Safety; Office of Railroad Safety of the Federal Railroad Administration
• Summary: The Federal Railroad Administration is the primary target entity that must conduct the study and publish results. The bill also identifies beneficiaries of the study results, including governmental entities regulating rail crossing safety and private entities maintaining rail crossing safety technology.

Enforcer

• Entity Types: AI_Governance_Actor
• Summary: No explicit enforcement body is named in the bill. However, Congressional oversight and general federal administrative law enforcement mechanisms would implicitly apply to ensure the Federal Railroad Administration complies with the statutory mandate.

Monitor

• Entity Types: AI_Governance_Actor
• Summary: No explicit monitoring body is designated. The public publication requirement creates transparency that enables oversight, but no specific entity is named to monitor compliance or implementation.

AI Lifecycle Stages

Lifecycle stages covered by the documents, as defined in the OECD Framework for the Classification of AI Systems

• Plan and Design
• Verify and Validate
• Deploy
• Operate and Monitor

AI Lifecycle Summary

The document addresses deployment and operational aspects of AI-enabled sensors at rail crossings by mandating a study that examines existing pilot programs, deployments, implementation best practices, and cost-benefit analyses. The focus spans planning considerations through operational deployment and monitoring of these safety systems.

Technical Scope

Terms relating to technical scope that are mentioned or implicitly referenced in the document

• AI Systems

Technical Scope Summary

The document explicitly references 'sensors enabled with artificial intelligence' throughout, treating these as AI systems for safety applications. However, it does not distinguish between different types of AI (frontier, general purpose, task-specific, etc.) or mention technical specifications like compute thresholds or model architectures.