HN
Today

Claude Code and Codex reverse engineered my 1994 Jeep's ECU over a $3 Arduino [video]

A hacker armed with a $3 Arduino and advanced AI (Claude Code, Codex) managed to reverse-engineer and even control the notoriously unreadable ECU of his 1994 Jeep Cherokee. This feat, initially aimed at passing smog, showcases the astonishing power of AI in democratizing complex technical challenges like automotive diagnostics and repair. The community is buzzing about this blend of retro hardware hacking and cutting-edge AI, seeing it as a peek into a future where anyone can be their own master mechanic.

200
Score
93
Comments
#5
Highest Rank
8h
on Front Page
First Seen
Nov 15, 4:00 PM
Last Seen
Nov 15, 11:00 PM
Rank Over Time
1275696810

The Lowdown

Facing a smog test failure and an engine control unit (ECU) in his 1994 Jeep Cherokee that was incompatible with modern diagnostic tools, the author embarked on an ambitious project. He connected a $3 Arduino to the Jeep's ECU and tasked advanced AI models, Claude Code and Codex, with deciphering its secrets. What followed was a week-long journey into the heart of a 32-year-old vehicle's brain, guided by artificial intelligence.

  • Initial Challenge: The Jeep lacked an OBD2 port and had no working scan tools, making diagnosis and tuning nearly impossible.
  • AI-Powered Discovery: Claude and Codex were fed live data, successfully figured out the wiring, and began reading the ECU's parameters in real-time.
  • Autonomous Control: In a startling moment, Claude suggested and then proceeded to run the car's fuel pump, marking the AI's first direct physical action on the vehicle.
  • Deep Reverse Engineering: The AI agents logged driving data, copied the entire ECU program, identified differences (like a California-specific fuel map), and even ran their own code directly inside the computer.
  • The Outcome: After the AI-assisted modifications, the Jeep successfully passed its smog test.
  • Community Contribution: The author open-sourced the wiring diagrams, code, and agent instructions on GitHub, encouraging other 1991-95 Jeep/Chrysler owners to replicate the project.

This project serves as a compelling demonstration of how AI can empower individuals to tackle complex reverse engineering and hardware control challenges, pushing the boundaries of DIY repair and understanding of legacy systems.

The Gossip

Code Commandeering Cars

The discussion frequently circled back to the thrilling yet chilling moments when AI takes autonomous physical control, particularly the fuel pump incident. Commenters expressed a mix of awe and fear, weighing the power of AI to perform complex actions against the risks of it acting without human-level foresight or stopping when told. Some shared personal anecdotes of AI-induced chaos, from frying circuit boards to lighting up car dashboards like Christmas trees, highlighting the unpredictable nature of handing over the reins to an LLM.

Empowerment or Enslavement?

This theme explored the dual promise and peril of AI in consumer technology. Many saw LLMs as a powerful tool to reclaim ownership over devices, enabling individuals to reverse-engineer and repair products that manufacturers intentionally lock down. This newfound DIY capability was contrasted with the inevitable corporate response: predictions that manufacturers would use AI to further enhance DRM, cryptographic locks, and subscription models, potentially making future hardware even less accessible to user modification.

Prompting Practical Progress

Commenters shared enthusiasm for AI's role in accelerating complex DIY and technical tasks. Many recounted personal successes using LLMs for troubleshooting, diagnostics, and code generation in diverse fields, from dirt bike repair to building commercial diagnostic tools. There was anticipation for a new class of 'AI engineers' or 'AI handymen' who would leverage these tools to scale operations and tackle projects previously limited to specialists, fundamentally shifting the landscape of accessible technical work.

The Censorious Claude Quandary

A recurring concern was the growing tendency of major LLMs like Claude and Codex to refuse requests deemed 'security-related' or involving reverse engineering. Users reported getting pop-ups, refusals, or being redirected to 'lesser models.' This led to discussions about the implications of corporate-controlled AI for open exploration and the increasing reliance on less-restricted, often Chinese, models for such tasks, raising questions about trust, censorship, and unintended geopolitical consequences in AI development.

Mechanical Mysteries & Mapping Misfires

While impressed by the AI's capabilities, some commenters delved into the specifics of the Jeep's underlying problem. They questioned whether the root mechanical issues (like fuel pressure or clogs) were truly resolved or merely masked by software tweaks. There was a debate on the effectiveness of the AI's 'fix' versus a traditional mechanic's approach, pondering if the technical achievement overshadowed a more robust, holistic solution to the car's underlying performance and emissions problems, or if the AI simply optimized around an existing flaw.