AI Take The Wheel: Driver Uses Claude to Hack a 1994 Jeep into Passing Smog Test
A Jeep enthusiast has successfully leveraged artificial intelligence to force a stubborn 1994 Jeep Cherokee to pass California's notoriously strict smog check after it failed due to high nitrogen oxide (NOx) emissions at 25 mph. Rather than turning to a mechanic or investing in expensive dealer diagnostics, Muhammad Zulali turned to a DIY approach involving an Arduino Nano, Claude, and Codex to bridge the gap between vintage hardware and modern AI.
The challenge lay in the vehicle's age. The 1994 Cherokee predates the OBD II interface standard, meaning its Electronic Control Unit (ECU) is completely alien to modern diagnostic tools. Unlike modern vehicles that can receive wireless patches or firmware updates, this vintage system required a much more invasive approach. Using approximately $15 in parts, Zulali connected an Arduino Nano directly into the Jeep's Chrysler SCI diagnostic port, effectively creating a bridge for the AI to interact with the vehicle's internal communication.
Reverse Engineering via AI
Once connected, Claude took over the software-side heavy lifting. The AI was able to decipher the esoteric baud rates and the inverted electrical signaling used in the car's communication protocol. This allowed the system to read live engine sensors and log critical data during test drives. As Zulali documented the project, he noted that while he handled the physical wiring and the actual driving, the AI managed the complex code analysis and memory manipulation.
The process went beyond simple data logging. During the tinkering phase, Claude successfully diagnosed low fuel pressure and issued a command to remotely fire up the vehicle's fuel pump. Because the 1993-era chip could not be simply reflashed like a contemporary ECU, OpenAI's Codex was utilized to help inject custom logic into the vehicle's EEPROM—the non-volatile persistent memory that stores critical configurations even when the battery is disconnected.
Results and Open Source Contribution
The path to success was not immediate. After multiple trial-and-error runs—where initial tweaks actually worsened emissions before being successfully dialed in—the 32-year-old Cherokee finally met the requirements to pass the smog test. Following this success, Zulali has open-sourced the utility code on GitHub under the name xj-scanner, providing a tool for other automotive enthusiasts to peer into the logic of legacy automotive brains.