AI Hijack Risk: New Vulnerability Found in Self-Driving Cars

New Cybersecurity Vulnerability Poses Risk to Self-Driving Vehicle Safety

A recently discovered cybersecurity vulnerability, dubbed VillainNet, could allow malicious actors to silently hijack the artificial intelligence (AI) systems controlling self-driving cars, raising significant concerns about the security of these increasingly prevalent autonomous vehicles. The vulnerability, identified by researchers at Georgia Tech, has the potential to compromise passenger safety and vehicle control.

The core of the problem lies in the architecture of the AI “super networks” that power modern autonomous driving systems. These networks are designed for flexibility, swapping out specialized subnetworks – essentially, AI “tools” – as needed to handle different driving conditions. According to David Oygenblik, a PhD student at Georgia Tech and lead researcher on the project, this adaptability, while beneficial for performance, creates a potential backdoor for attackers.

“Super networks are designed to be the Swiss Army knife of AI, swapping out tools, or in this case sub networks, as needed for the task at hand,” Oygenblik explained. “However, we found that an adversary can exploit this by attacking just one of those tiny tools. The attack remains completely dormant until that specific subnetwork is used, effectively hiding across billions of other benign configurations.”

How VillainNet Works

VillainNet operates by embedding itself within one of these subnetworks. It remains inactive until the conditions are right for activation – for example, when a self-driving taxi’s AI responds to changing weather conditions like rainfall. Once triggered, the vulnerability is highly likely to succeed in granting the attacker control of the vehicle. Researchers demonstrated a 99% success rate in activating the attack while maintaining its stealth throughout the AI system.

The potential consequences of a successful attack are alarming. Hackers could, in theory, hold passengers hostage or even threaten to crash the vehicle. The researchers emphasize that the attack isn’t limited to causing immediate accidents; it could be used to subtly redirect a vehicle to an attacker-chosen destination without the occupants even realizing they are off course. This is particularly concerning for passengers unfamiliar with the area.

The Challenge of Detection

What makes VillainNet particularly dangerous is its elusiveness. Current cybersecurity tools struggle to detect the vulnerability. Oygenblik notes that finding and neutralizing VillainNet is akin to “finding a single needle in a haystack that can be as large as 10 quintillion straws.”

The research indicates that verifying the safety of an AI system against a VillainNet-style attack would require 66 times more computing power and time than current methods allow, making comprehensive security checks impractical. Which means the vulnerability could be hidden at any stage of development and remain undetected for extended periods.

Beyond Immediate Crashes: A New Threat Landscape

This discovery builds on growing concerns about the cybersecurity of autonomous vehicles. Previous research has focused on attacks that cause immediate safety failures, such as crashes or traffic violations. However, VillainNet represents a qualitatively different risk – a long-term compromise of route integrity.

Recent work, including a study highlighted by February 6, 2026 research published on arXiv, demonstrates that attackers can subtly manipulate a vehicle’s trajectory over time, steering it away from its intended route. This type of attack, while not immediately dangerous, could have serious implications for passenger safety and security.

Industry Response and Future Security Measures

The Georgia Tech researchers presented their findings at the ACM Conference on Computer and Communications Security (CCS) in October 2025, issuing a “call to action” for the security community. They suggest that bolstering the security of these AI super networks is crucial. The hypothetical solution involves adding security measures to the networks themselves, recognizing that the vulnerability stems from the flexibility that allows for the swapping of subnetworks.

The increasing complexity of AI systems necessitates the development of new defenses capable of addressing these novel, hyper-targeted threats. As autonomous vehicles become more integrated into our transportation infrastructure, ensuring their cybersecurity will be paramount to public safety and trust. Waymo, a leading developer of self-driving technology, already emphasizes the importance of cybersecurity, stating on its website that “protecting the Waymo driver from malicious activity is paramount” and that they have developed a “robust process to identify, prioritise, and mitigate cyber security threats.”

The research underscores the need for ongoing vigilance and innovation in the field of autonomous vehicle cybersecurity. The potential for malicious actors to exploit vulnerabilities in these systems is real, and proactive measures are essential to protect passengers and the public.