AI
The EV Charger Attack Surface
EV chargers run a main CPU on Linux or Android, often a second processor for power switching, and connect over Wi-Fi, Bluetooth, cellular, and multiple cloud protocols. A companion mobile app delivers Wi-Fi credentials over Bluetooth, then handles telemetry and control. That app is free to download and reveals substantial detail about how these systems work.
Network interfaces, cloud back-ends, the mobile app, and the physical connectors all add up to a large attack surface before a single exploit is written. Jonathan Andersson and Thanos Karageorgis documented this surface across eight chargers that have passed through Pwn2Own competitions.
Pwn2Own Results: Bug Classes Across Eight Chargers
Between 2024 and 2025 Pwn2Own Automotive, contestants found buffer overflows (CWE-120/121/122), authentication failures, and command injection across eight EV chargers — all simple, unchained bugs. Senactive downgraded a Tesla Wall Connector to a debug firmware, extracted Wi-Fi credentials, then hit a global buffer overflow with no memory protections for immediate RCE. The Wolf Box shipped a Mifare Classic 1K RFID card that resets every Wolf Box ever made; cloning it forced re-pairing, leaking a firmware key used to push malicious firmware. The worst: a ChargePoint cloud bug granted fleet-wide firmware write access and was still unpatched at the time of the talk.
How the Charging Protocol Carries Current — and Risk
The J1772 standard covers Level 2 charging up to 80 amps. Inside the charging gun, a 1 kHz PWM signal called the control pilot tells the car its maximum allowed draw: 10% duty cycle means 6 amps, 96% means 80 amps. An attacker who controls the charger’s processor controls that signal directly. At Pwn2Own, contestants demonstrated live manipulation of the duty cycle after compromising a charger, varying the current in real time.
The follow-on research question: if no software safety fires and the signal is pegged at maximum, what does the hardware do?
Building a Test Rig to Stress Compromised Chargers
To answer that question, the team built a 27.12-kilowatt resistive load bank: a breaker, the charger under test, a J1772 load controller with remotely-switched heaters, and dual current readouts. Remote Wi-Fi breakers kept hands clear of live conductors.
For the software bypass, they reverse-engineered each charger to find the processor pin driving the main relay, lifted it, and ran two traces to external headers. A jumper reconnects the net for baseline tests; removing it gives the researchers direct relay control, bypassing all software safeties while leaving any downstream hardware protections intact. They ran this procedure on eight designs across five days.
What Actually Happens When a Compromised Charger Pushes 80 Amps
During the compromised charger thermal and arc flash test (21:34), 78 amps through an Autel charger hit 340°F in 25 minutes and melted the cable holder off the wall. A second device ignited its own toxic, flammable fumes twice. Half the chargers failed within 90 minutes; the last lasted five and a half hours. In three cases, the melting cable shorted high current into the control pilot conductors, potentially damaging the connected EV.
“none of the chargers we tested were robust in the face of compromise” — Jonathan Andersson
Why Circuit Breakers Don’t Solve This — and What Would
A 60-amp circuit breaker does not trip the instant you exceed 60 amps. A 48-amp charger installs on a 60-amp breaker by code, and that breaker has thermal headroom on top.
“a 48 amp charger can overcurren by 41% indefinitely” — Jonathan Andersson
The fix is a hardware current transformer placed downstream of the processor’s last control point, with a set-point fixed at manufacturing time. Software cannot reach it. A latched interrupter drops the relay when current exceeds the limit. A fuse achieves the same result. Neither is expensive. No tested charger had either.
Notable Quotes
none of the chargers we tested were robust in the face of compromise Jonathan Andersson · ▶ 27:52
a 48 amp charger can overcurren by 41% indefinitely Jonathan Andersson · ▶ 29:29
under no circumstance should a device continuously draw, you know, multiv multiples of its maximum rated current Jonathan Andersson · ▶ 34:56
Key Takeaways
- Eight EV chargers tested at Pwn2Own 2024-2025 fell to single-step exploits including buffer overflows and a fleet-wide cloud bug.
- No tested charger had hardware overcurrent protection independent of software; all failed dangerously within five and a half hours.
- A hardware current transformer set at manufacturing time, outside any software path, would prevent physical destruction even on a fully-compromised charger.