This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
NJ Transit relied on National Weather Service storm modeling software to decide where to store rail equipment ahead of Hurricane Sandy. By inputting incorrect storm trajectory and speed data, the agency underestimated the potential storm surge, leading to the flooding of the Meadows Maintenance Complex and $100 million in equipment damage. Experts and federal meteorologists confirmed that the software was used incorrectly, and that proper inputs would have indicated a high risk of inundation.
New Jersey Transit's use of a federal government storm modeling software underestimated the threat of storm surges to the Meadows Maintenance Complex, leaving millions of dollars worth of equipment in the rail yard before Hurricane Sandy struck.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Users anthropomorphizing, trusting, or relying on AI systems, leading to emotional or material dependence and inappropriate relationships with or expectations of AI systems. Trust can be exploited by malicious actors (e.g., to harvest personal information or enable manipulation), or result in harm from inappropriate use of AI in critical situations (e.g., medical emergency). Overreliance on AI systems can compromise autonomy and weaken social ties.
NJ Transit officials relied on the outputs of the SLOSH simulation software despite inputting incorrect parameters and ignoring explicit, severe storm surge warnings from the National Weather Service.
Human
Due to a decision or action made by humans
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was caused by human operators inputting incorrect parameters into the SLOSH software, leading to an unintended underestimation of the storm surge during active deployment.
NJ Transit staff incorrectly configured the NWS SLOSH storm surge model, leading to an underestimation of Hurricane Sandy's impact. Relying on this flawed output, the agency stored locomotives in low-lying yards, resulting in over 100 million dollars in damage to transit infrastructure and months of severe commuter disruptions, highlighting critical infrastructure vulnerabilities to model overreliance.
The SLOSH software operates as a Tool, performing the specific function of simulating storm surges based on direct user inputs (such as storm speed and trajectory) to produce deterministic outputs.
High Risk: The system is used in critical infrastructure, specifically for emergency planning and decision-making in public transit (NJ Transit), which has significant implications for public safety and infrastructure protection.
The SLOSH software is designed to model and simulate storm surge behavior to estimate potential flooding heights and detect threats to coastal areas.
The reports provide clear, consistent, and highly detailed accounts of the incident, including specific technical details about the incorrect inputs (northeast trajectory, 10 mph speed), the resulting underestimated surge (3 feet vs 9.23 feet), and the exact financial and physical damage ($100-$120 million, over a quarter of the fleet). Statements from NWS meteorologists and NJ Transit officials are directly quoted, leaving very little ambiguity about the sequence of events.
One could argue this is purely a human management failure rather than an AI safety incident, but the central role of the SLOSH simulation software in justifying the decision makes it a clear case of overreliance on incorrectly used technology.
Technology
Incorrect Simulation Inputs
Software modeled wrong storm direction and speed, underestimating surge.
Misuse of SLOSH Software
NWS simulation software was used incorrectly by untrained transit officials.
Lack of Software Guardrails
The software allowed highly unrealistic inputs without flagging anomalies.
Data Inputs
Flawed Weather Model Inputs
Inputs had storm going northeast at 10mph instead of west-northwest at 20mph.
Ignored Official Forecast Data
Official NWS warnings of 15-foot surges were not used in the local model.
Lack of Real-Time Data Validation
No process to validate simulation inputs against actual NWS briefing data.
Human Factors
Confirmation Bias in Modeling
Officials accepted low surge outputs because they aligned with history.
Lack of Technical Expertise
Staff lacked training to correctly operate the storm simulation software.
Overreliance on Flawed Model
Blind trust in the incorrect 3-foot surge output over actual warnings.
Process and Methods
No Consultation with Experts
NJ Transit failed to contact NWS to verify their simulation parameters.
Inadequate Emergency Modeling
Modeling was treated as a real-time decision tool rather than planning tool.
Lack of Peer Review for Models
No internal or external review of the simulation setup was conducted.
Management
Ignored Climate Risk Reports
Leadership failed to study or act on internal climate vulnerability reports.
Poor Leadership Communication
NJ Transit executive director and Governor did not communicate before storm.
Security Prioritized Over Weather
Emergency planning focused on active shooters rather than weather events.
Regulatory Environment
No Standard for Transit Modeling
Absence of regulatory standards for climate risk and storm surge modeling.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
