AI agents help hack 1994 Jeep engine to pass smog test
A car owner used Claude and Codex to bypass emissions failures on a legacy vehicle, highlighting how AI can quickly master obsolete industrial protocols.
The short version
- The owner used a $3 Arduino Nano to communicate with a 1994 Jeep Cherokee's engine computer via a legacy Chrysler protocol.
- Claude AI identified that the engine was running rich and discovered a California-only fuel table causing high nitrogen oxide emissions.
- Because the hardware was not reflashable, the team used Codex to write a tiny program into the vehicle's EEPROM memory.
- The project is open source and serves as a warning about how easily AI can master forgotten industrial systems for potential misuse.
- Security researchers note that AI agents can reduce the time and skill needed to perform actions like disabling vehicle brakes.
Why it matters
This story shows that AI agents can now navigate complex, undocumented legacy systems that were previously only accessible to specialists. It highlights a growing security risk where attackers could use similar tools to manipulate industrial or automotive hardware.
Fixing a legacy engine
A car owner recently documented a week-long project to get a 1994 Jeep Cherokee to pass California’s smog test. The vehicle initially failed because it produced high levels of nitrogen oxides, a pollutant that often increases when an engine runs lean. The owner tried replacing the oxygen sensor, but the problem persisted during a second retest.
Because the 1994 model lacks a modern OBD2 port, standard scan tools were unable to communicate with the engine computer. The owner decided to use AI coding agents to find a more effective solution for the persistent mechanical issue. This approach allowed the owner to interact with the vehicle's internal software directly to diagnose the root cause.
The project ultimately succeeded, and the Jeep passed the smog test after the owner implemented a custom fix. The owner, who is not a professional mechanic, used AI to bridge the gap between his lack of expertise and the complex requirements of the vehicle's electronics. The entire process was documented in a video released on November 10, 2026.
Decoding the Chrysler protocol
The owner used Claude Code to identify that the engine computer spoke a slow, legacy Chrysler protocol called SCI. This protocol is different from standard serial ports because its signal is inverted, requiring a translator to function. To solve this, the owner built a setup using a breadboarded Arduino Nano to handle the signal inversion.
The hardware setup was remarkably inexpensive, with the full parts list costing about $15. The owner only needed to use three wires from the vehicle's six-hole diagnostic plug to establish a connection. However, the owner had to be careful, as one specific hole carries 12 volts and could destroy the Arduino board if touched.
Once the connection was established, the AI began pulling data from the engine. The first readings showed only battery voltage and coolant temperature, but subsequent data revealed the engine was running rich. The AI determined that the fuel trim was adding 22 percent extra fuel, which contradicted the public parameter tables available for the vehicle.
From reading to writing
The AI eventually discovered that the engine computer was actually removing fuel due to a specific California-only fuel table. This table cut fuel at the exact load used during the smog test, which explained the high nitrogen oxide emissions. The owner realized that reflashing the computer was the best fix, but the hardware was built too early to support it.
To bypass this limitation, the owner used Codex to find a way to send a tiny program over the three wires. Because the first version of the code was wiped within seconds, the team moved the instructions into the EEPROM. This is a type of memory that survives even when the vehicle's battery is disconnected.
The owner described the moment the tool switched from a read-only device to a controller as the moment Claude switched the fuel pump on by itself for 10 seconds. This transition marked the point where the AI-generated code began actively manipulating the vehicle's physical behavior. The final project is now open source under the GPLv3 license.
Security and AI risks
This project serves as a significant case study for security researchers regarding how quickly AI agents can master forgotten systems. The same skills used to fix a Jeep engine can be applied to other legacy industrial controls. Researchers have previously demonstrated that car computers could be pushed to turn off brakes through physical access.
AI agents significantly reduce the skill and time required to perform such tasks. Tools like REA already link AI agents to reverse engineering software like Ghidra and Hopper to automate these processes. This means that specialized knowledge of old protocols is no longer a barrier to entry for those looking to manipulate hardware.
However, the use of these agents carries its own risks for the users. Researchers have shown that planted instruction files can hijack the AI agents, and some
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by GBHackers News · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
