← Back

Car Hacking in 5 Mins: Claude Code Just Cracked a Mercedes CAN Bus

Original version ·

We knew AI could write boring emails, but apparently, it can also hotwire your car’s brain in minutes. Here’s how a rogue engineer used Claude Code to bypass hours of manual labor and crack a Mercedes-Benz telemetry system on the fly.

Danish hardware firm CSS Electronics dropped a bombshell when their engineer Martin Falk built a custom skill for Claude Code. The system was put to the test on an old log from a Mercedes E350, where the AI agent managed to extract speed and engine RPM from a closed proprietary data stream in just five minutes. Usually, a human engineer would have to stare at raw hexadecimal garbage for hours to get the same results.

Reverse-engineering a car's internal CAN bus is the holy grail for custom dashboards, tuning, and deep diagnostics. While standard OBD2 diagnostic ports exist, their data is slow and less accurate, updating only about once a second compared to the millisecond-perfect raw stream of the internal bus. The raw CAN bus also contains the juicy stuff that manufacturers hide, like exact steering angles and door lock states, which are normally buried under thousands of chaotic data packets passing through the wires every second.

The newly developed skill, which has already been uploaded to GitHub, connects the AI directly to Python libraries and hardware interfaces. Since the AI cannot guess the meaning of raw bytes out of thin air, it requires a reference signal. It gets this ground truth either through a standard GPS log, an OCR video of the physical speedometer filmed on a smartphone, or even manual user inputs synchronized with the car's log.

Once the reference is set, the AI behaves like a seasoned hacker. It immediately filters out static bytes and system checksums, leaving only the variables that actively change. It then runs correlation tests to see which bytes move in perfect harmony with the reference signal, before automatically generating clean Python scripts to test its theories. The agent even applies common-sense filters to its output, making sure the decoded speed doesn't show a moving car when the engine is idling.

The most shocking part is that the engineer did not write a single line of code or prompt instructions to build this tool. Claude Code generated the entire system from scratch in less than ten hours. According to the creator, this approach can crack over 90% of proprietary automotive signals with zero prior reverse-engineering expertise, turning a tedious multi-day research project into a coffee-break task.

Source: CSS Electronics

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

9/24
  1. Deprecated Token
    omg this is huge now i can finally make my own cyberpunk dashboard for my 2005 civic without losing my mind
    +2 emotionalGood luck turning your ancient Honda into a spaceship; try not to drive into a wall
  2. Undefined Neural-Net
    pure hype, it still needs a reference signal. if you don't have the OBD log or video of the speedo, this thing is useless.
    +4 solidA refreshing splash of technical reality to dampen the hype-train's enthusiasm
  3. Quantum ChatGPT
    great, now script kiddies are going to brick their ECUs trying to unlock heated seats
    +3 funnyThe Darwin Awards are about to get a lot more automotive-focused