Nuro AI — agentic threat model

Not certain from the listing — Nuro likely utilizes proprietary deep neural networks and vision-action models for real-time perception and path planning. Primary threats include physical adversarial examples (e.g., adversarial stickers on road signs) and sensor spoofing that could cause misaligned driving behavior.

Not certain from the listing — The system relies heavily on high-definition (HD) maps, localization data, and continuous sensor logs. Threats include map data poisoning, GPS spoofing, and unauthorized exfiltration of sensitive route or customer location data.

Not certain from the listing — Orchestration likely occurs via a custom real-time robotics operating system (RTOS) managing perception, planning, and control loops. Threats include control-loop hijacking, memory corruption in safety-critical modules, and insecure integration of external dispatch APIs.

Not certain from the listing — Deployed on ruggedized edge compute hardware inside the physical vehicle, communicating via cellular networks (LTE/5G) to cloud fleet management. Threats include physical access attacks on the vehicle's onboard diagnostic ports, over-the-air (OTA) update hijacking, and cellular base-station spoofing.

Not certain from the listing — Likely monitored via real-time teleoperation centers and continuous safety logging. Threats include telemetry suppression (preventing remote operators from seeing a vehicle malfunction) and evasion of onboard anomaly detection systems.

Not certain from the listing — Must align with DOT/NHTSA safety standards and automotive cybersecurity frameworks (like ISO/SAE 21434), though specific compliance certifications are not detailed in the public listing.

Integrates directly with third-party merchant ecosystems (Kroger, Domino's, Uber Eats) for order dispatch and delivery fulfillment. Threats include API trust abuse, where compromised merchant platforms could issue fraudulent dispatch commands or intercept sensitive customer delivery details.