AI and Trucks: What Could Go Wrong?

Celeste Polidori
April 14, 2025

Sudden route changes, ethical dilemmas—like a dog on the road—or GPS failures cast doubt on putting full trust in these vehicles. These are the issues that have raised red flags.

The development of autonomous trucks, powered by artificial intelligence systems, is moving rapidly across the globe.

Companies like Waymo, Tesla, TuSimple, and Aurora are testing and deploying vehicles on real-world routes, with the promise of transforming freight transportation.

However, this type of technology carries a number of risks, some already documented in specific situations.

Software Failures or System Malfunctions

In 2023, a self-driving truck from TuSimple made an unexpected maneuver on an Arizona highway: it abruptly turned left and crashed into a barrier. The company blamed the incident on human error in the interface, but internal documents revealed that the system had misinterpreted a remote command due to synchronization issues between sensors and controls.

Errors in Environmental Perception

Sensors can fail to correctly identify elements in the environment. In 2018, a self-driving Uber vehicle, during testing in Tempe, Arizona, struck and killed a woman crossing the street at night with her bicycle. The sensors detected the object, but the system failed to properly classify the threat and didn’t activate emergency braking.

Decision-Making in Ethical Dilemmas

Situations like deciding whether to swerve for a stray animal in the road remain unresolved challenges. While there’s no public case involving a truck, classic dilemmas—“run over the dog or swerve and risk a crash?”—are part of the simulations AI programmers face. These decisions are hard-coded and can’t be ethically adapted in real time.

Unexpected Obstacles Not Covered in Training

In 2022, a self-driving truck from Embark came to a full stop on a highway after encountering a misplaced traffic cone. Since the system hadn’t been trained on how to react in that specific scenario, it chose to stop completely, blocking traffic until human assistance arrived.

Cyberattacks and Network Vulnerabilities

In controlled environments, security researchers have successfully hacked autonomous vehicles and altered their behavior.

In 2015, hackers publicly demonstrated how they could remotely control a moving Jeep Cherokee—changing its speed and direction. Although it wasn’t an autonomous truck, the incident exposed real vulnerabilities in connected vehicles.

Mismatch Between Autonomous and Human-Driven Vehicles

Humans can behave unpredictably, and autonomous systems struggle to anticipate certain maneuvers. In San Francisco, several Cruise autonomous cars (a GM subsidiary) got stuck at an intersection during a large public event, unable to correctly interpret temporary detours set by human police officers.

GPS Signal Loss or Geolocation Errors

In tests conducted in rural areas or tunnels, various manufacturers have documented temporary GPS losses. Some systems respond by stopping for safety, which can be dangerous on high-speed roads if there’s no shoulder or proper visibility.

Impact of Sudden Road Changes

On poorly maintained roads, where potholes or debris appear without warning, the system may not react in time. While these events haven’t led to serious reported accidents involving autonomous trucks, there have been instances of emergency braking, posing risks to vehicles traveling behind.

Ambiguity in Human Signals

In multiple user-recorded videos, test vehicles have been seen struggling to interpret informal traffic gestures: a traffic officer signaling without a light, or a pedestrian waving a vehicle through. Current systems are not well-equipped to handle this kind of ambiguity.

Dependence on Remote Connectivity

Many systems rely on constant communication with servers or monitoring centers. In areas with limited or unstable data coverage, that connection can be lost. Some companies implement “safe modes” that stop the truck, but even that stop can pose a risk if it happens on a busy highway.