Here’s How AI Agents Can Protect EV Chargers

The number of electric vehicles on roads around the world continues to grow. The boom in EV adoption has driven the development of accessible, fast, and efficient charging infrastructure.

However, this expansion also brings with it new cybersecurity risks that have been not been widely studied, and for which there are still few viable solutions.

Cristina Alcaraz, an infrastructure-security researcher at Spain’s University of Malaga, explains that the liability of electric-vehicle charging stations is due to the fact that they integrate multiple physical and digital components. She says this complex architecture not only keeps the chargers working efficiently but also presents a host of new and far-reaching security vulnerabilities. Chargers’ exposure to attacks compromises both the continued adoption of EVs as well as the stability of the electrical grids in the countries where chargers operate.

With the aim of tackling this threat, researchers from the NICS lab at the University of Malaga have developed an innovative proposal to deploy AI agents to protect the infrastructure. These agents are designed to prevent cyberattacks from different vectors, ranging from fraud or energy theft by malicious actors using the charging stations to larger attacks that could damage critical-energy networks.

The team’s proposal aims to ensure the early and reliable detection of anomalies and attacks to charging networks using the Open Charge Point Protocol. The OCCP standard is one of the most widely used for the operation and management of electric-vehicle chargers. The protocol allows a network of charging stations to communicate with a centralized system that can manage, monitor, and coordinate all energy transactions carried out by the end users.

The central system handles a bunch of things remotely, including user authentication, management of the electrical load at each station, monitoring of overall electricity consumption, and technical diagnostics. These capabilities allow for real-time infrastructure control and enable operators to spot and rapidly respond to any anomalous behavior.

However, the authors of the new study point out that current monitoring mechanisms based on this protocol typically just focus on network traffic or local events, so they can only offer a limited view of what is happening across an entire region of infrastructure. The researchers say this limitation makes it difficult to identify where in the system an anomaly is occurring, which network components are compromised, the extent of any vulnerabilities, and the ways in which a potential attack might spread.

Call in the AI

The researchers propose a system that uses multiple AI agents. Each station or relevant component of the charging network incorporates AI agents that are capable of analyzing their environment, collecting information, and collaborating with other agents in order to build a comprehensive view of the infrastructure’s present state.

“Each agent assesses the status of chargers, communications, and connected devices to detect anomalies, operational failures, or potential security incidents,” says Alcaraz. “These agents, which are connected to a central-monitoring system, compare the information obtained locally with that of nearby stations, providing a more complete, accurate, and contextualized collaborative view of the situation,” she says. Alcaraz is also the lead author of the report.

The work, published in the International Journal of Critical Infrastructure Protection, explains that one of the most novel features of the system is its use of a consensus mechanism based on a mathematical framework known as opinion dynamics.

This approach mimics the processes by which humans exchange information within their own social networks to reach agreements. When applied to computer models, it allows AI agents to share observations with each other and gradually adjust their assessments to build a collective understanding of the overall situation.

According to the authors, this procedure reduces the risk of the AI agents generating false positives. It also lets the system detect anomalies that might go unnoticed if they were only analyzed locally.

The proposed architecture also uses blockchain tech as a trust and validation mechanism. All transactions performed by the agents are recorded in a distributed ledger that cannot be altered afterward, guaranteeing the system’s integrity and traceability.

Stress Test

A multi-agent system was tested by researchers in a simulated OCPP-compliant charging environment. During the experiments, the agents were exposed to various anomaly scenarios within the charging network: component failures, communication link errors, and situations that required a coordinated response from multiple parts of the system. In all cases, the AI agents had to identify each local disturbance, share their observations with each other, and collaborate to build a shared understanding of the incident.

The results showed that the combination of the AI agents, the distributed-consensus mechanism, and blockchain technology provided a global view of the network. The system detected both specific anomalies in individual devices and some behavioral patterns that were affecting multiple-charging stations. Furthermore, the consensus mechanism improved the accuracy of the diagnoses by comparing observations from different agents, increasing the reliability of the reports.

The university lab is pleased with the results. “This system provides a new way to guarantee the protection of electric-vehicle charging infrastructure,” it said in a press statement.