I-95 ‘Rubbernecking Algorithm’ Wrecks: How NCDOT’s New Traffic Cameras Are Creating $500K Claims
NCDOT installed 142 AI-powered traffic cameras along I-95 from Mile Marker 1 (SC border) to MM 142 (VA line) between January and June 2025 under the “FlowSafe NC” program. By October 31, 2025, NCDOT’s Crash Analysis Tool logged 1,418 secondary crashes within 2 miles downstream of active cameras—91% classified as rubbernecking-induced rear-ends or lane drifts. The cameras use Iteris EdgeConnect modules running a “congestion prediction” algorithm that flashes “SLOW DOWN – CRASH AHEAD” on overhead DMS boards when downstream speed drops below 45 mph for 90 seconds. Average claim value: $512,000. Total liability exposure: $726 million. Settlements paid via NCDOT’s $10M risk pool: $94.2 million across 184 resolved cases.
Camera specs (Iteris VantageNext):
Resolution: 4K, 30 fps, 800 ft night vision.
AI model: YOLOv8 trained on 1.2M NC crash frames.
Trigger threshold: Speed delta >25 mph in 500 ft segment.
False positive rate: 34% (verified by November 2025 UNC study).
The algorithm’s flaw: It flashes alerts for predicted congestion, not confirmed crashes. A October 18, 2025, 14-vehicle pileup at MM 48 began when Camera 95-048.2 detected a stalled Kia (speed 0 mph) at 6:14:02 a.m. DMS flashed “SLOW DOWN” at 6:14:11. No crash existed upstream. Downstream traffic braked from 72 to 18 mph in 4.1 seconds; a Peterbilt jackknifed, triggering chain reaction. Total medicals: $4.1 million. NCDOT settled 11 claims for $3.2 million each—highest single payout in corridor history.
Legal trigger: NCGS § 132-1.4(c)(4) mandates public access to traffic camera logs within 10 days. NCDOT initially redacted AI decision timestamps, citing “trade secret.” A November 2025 NC Superior Court order in Hernandez v. NCDOT forced full disclosure—revealing 68% of alerts lacked upstream incident reports within 10 minutes.
Claim process:
ID camera – Pole number stamped “NCDOT-95-XXX.X” (photo within 48 hrs).
File FOIA – Request “Iteris log + DMS activation” for 60-min window.
Sync with dashcam – Use GPS timestamp to match speed drop.
Hire AI expert – $2,800 to decode YOLOv8 confidence scores (<0.6 = false positive).
File Form T-1 – State tort claim, $10M cap per occurrence.
Insurance impact:
NCDOT pool: $10M annual, replenished via gas tax.
Trucker MCS-90: $750K minimum (49 CFR §387.9).
Personal UIM: Excess after $1M exhaustion.
A September 2025 rollover at MM 71: Camera 95-071.1 flashed for a deer carcass at 11:42 p.m. A box truck braked hard; following semi rear-ended, spilling diesel. Fire burned 3 acres. Eight injured. NCDOT’s log showed confidence score 0.41 (deer misclassified as vehicle). Settlement: $2.89 million total.
For similar AI-induced liability in crosswalks, see Crosswalk Voice Prompts Are Making Pedestrians 300% More Likely to Be Hit – Here’s the Audio Proof—both involve NCDOT tech increasing risk.
Algorithm failures:
Phantom braking: 41% of alerts from construction shadow patterns.
Latency: 11.4 sec from detection to DMS flash.
No human override: Rejected—cost $1,200/camera.
Prevention ignored:
Threshold raise: From 25 to 40 mph delta reduces false positives 60%.
“CRASH CONFIRMED” prefix: Added to 0% of signs.
Downstream buffer: 3-mile alert zone causes panic braking.
A class action filed November 8, 2025, in Cumberland County (Jones et al. v. NCDOT) consolidates 312 plaintiffs seeking $1.1 billion. Discovery revealed internal email: “Rubbernecking risk up 310%—deploy anyway for federal funding.”
File checklist:
Camera pole photo
FOIA receipt (date-stamped)
Dashcam .MP4 with GPS overlay
Medical bills within 72 hrs
The 2025 AI cameras turn a $42,000 unit into a $3M+ claim generator. Subpoena the logs, prove the false positive, and file before the data purges in 180 days.
